Repository navigation
Examples
Every clip on this page was generated inside Forge Neo by this extension (native backend, versions 0.3.0 to 0.6.0), with picture and sound made together. Nothing was upscaled, retimed or edited afterwards: these are the MP4 files exactly as Forge saved them.
Click a preview to download the MP4 with sound. The previews are silent GIFs at a lower frame rate; the sound is a big part of these clips, so listen to the MP4.
Common setup: unless a section says otherwise, one NVIDIA A40 (48 GB) with about 50 GB of system RAM on RunPod, Forge Neo neo-2.29.2, 24 FPS, CFG 1, Simple schedule, seed fixed. Unless a clip says otherwise, the model set is the standard one from Getting Started: the Comfy-Org pruned INT8 checkpoint, the INT8 text encoder, the fp16 video VAE and the fp32 audio VAE. Times are the full request, as measured through Forge's API; when the model was not loaded yet, loading is included. A dash means the time was not recorded.
- Long clips (15 seconds)
- Starting from a picture
- People and speech
- Around the world
- Motion
- FastH3
- Community checkpoint: H3 Eros Max
- More from the author
- More clips on other pages
- All clips at a glance
A 15-second vertical clip in a frantic music-video style: about thirty shots with whip pans, crash zooms, flash transitions and a breakbeat track. The prompt is a structured shot list with timings. Made with the turbo LoRA at 8 steps, applied on the fly (see the memory note below). The same prompt with H3 Eros Max is further down.
| Model | Size | Frames | Sampler | Steps | Shift | Seed | Time |
|---|---|---|---|---|---|---|---|
| Standard + Comfy-Org turbo LoRA (weight 1) | 576×1024 | 362 (15.08 s) | Res Multistep | 8 | 6 | 864569260196868 | 840.8 s (14 min) |
Note
At 15 seconds the system RAM peaked at 46.3 GiB of the 46.6 available. With the turbo LoRA in Forge's default mode the process ran out of RAM; setting Diffusion in Low Bits to Automatic (fp16 LoRA) applies the LoRA on the fly instead of keeping a second copy of the weights, and the clip finished.
Prompt
subject_definitions:
<Subject 1> is a beautiful young Japanese woman with long, straight dark hair and fair skin, wearing a trendy street-style outfit consisting of a beige oversized blazer, a white crop top, and a pleated skirt. She is seated or standing at a modern urban bus stop.
summary:
[reference generation] The target video features <Subject 1> performing a series of cool and cute poses at a bus stop. The video strictly adopts the hyper-kinetic, disorienting editing style, utilizing extremely rapid cuts, aggressive motion-blurred transitions, extreme low angles, and frantic handheld camera movements to create a high-energy, music-video-like atmosphere.
detailed_description:
The target video utilizes a hyper-kinetic, "action cam" aesthetic with frantic handheld shaking, extreme Dutch angles, and aggressive motion-blurred transitions with chaotic editing rhythm.
[Shot 1] The video opens with a violent radial blur explosion, instantly snapping into focus on <Subject 1> sitting at the bus stop. The camera is positioned at an extreme low angle, tilted heavily to the right (Dutch angle).
[Shot 2] At 00:00.500, a fast whip-pan transition blurs the screen. The shot cuts to an extreme low-angle fisheye close-up of <Subject 1>'s white sneakers tapping on the pavement. The camera rapidly pushes in.
[Shot 3] At 00:01.000, a rapid upward whip-pan cuts to a tight close-up of <Subject 1>'s face. The camera is tilted 45 degrees. She flips her hair gracefully, looking away with a cool, aloof expression.
[Shot 4] At 00:01.500, a pure white overexposure flash transition wipes the screen.
[Shot 5] At 00:02.000, the camera performs a frantic handheld orbit around <Subject 1>. She rests her chin on her hand, looking bored. The background street traffic is blurred into light streaks due to the motion.
[Shot 6] At 00:02.500, a crash zoom aggressively pushes into her eyes. She suddenly blinks and breaks into a bright, cute smile.
[Shot 7] At 00:03.000, a rapid tilt-down shot focuses on her fingers tapping a rhythm on her pleated skirt.
[Shot 8] At 00:03.500, a heavy motion-blur lateral pan transitions to a side medium shot.
[Shot 9] At 00:04.000, an extreme low-angle shot looks up at <Subject 1>. She tilts her head down to look at the lens, her dark hair falling forward playfully.
[Shot 10] At 00:04.500, a rapid zoom-out reveals the full bus stop. <Subject 1> stands up, one hand in her pocket, striking a confident, cool pose.
[Shot 11] At 00:05.000, the camera shakes violently as if running, cutting to her back. She turns her head to look over her shoulder.
[Shot 12] At 00:05.500, a whip-pan snaps back to the front. <Subject 1> suddenly leans extremely close to the lens, making a cute "peace" sign.
[Shot 13] At 00:06.000, a radial blur transition spins the frame.
[Shot 14] At 00:06.500, a close-up of her profile. The wind blows her hair. The footage suddenly drops into slow-motion, then instantly speed-ramps back to hyper-speed.
[Shot 15] At 00:07.000, an extreme low-angle shot glides rapidly along the ground. <Subject 1> lifts her foot and taps the ground.
[Shot 16] At 00:07.500, a crash zoom pushes into a macro shot of her silver necklace and blazer collar.
[Shot 17] At 00:08.000, a Dutch-angled handheld shot orbits halfway around her. She spins, her skirt flaring out.
[Shot 18] At 00:08.500, a motion-blur transition cuts to a wide shot. A bus arrives in the background. <Subject 1> stands by the stop sign.
[Shot 19] At 00:09.000, an extreme low-angle wide-angle shot. She stands with hands on her hips, looking confidently into the distance.
[Shot 20] At 00:09.500, a rapid whip-pan to the bus window, briefly capturing her reflection.
[Shot 21] At 00:10.000, a radial blur explosion cuts back to her sitting on the bench, facing forward.
[Shot 22] At 00:10.500, a crash zoom pushes into her lips. She gently bites her lower lip, giving a charming gaze.
[Shot 23] At 00:11.000, a rapid zoom-out shows her stretching her arms above her head, showing a cute, relaxed side.
[Shot 24] At 00:11.500, the camera shakes violently, cutting to a close-up of her shoes scuffing the pavement.
[Shot 25] At 00:12.000, a whip-pan upward reveals her making a finger-heart gesture directly at the camera.
[Shot 26] At 00:12.500, a pure white overexposure flash transition.
[Shot 27] At 00:13.000, an extreme low-angle shot. She squats down, bringing her face extremely close to the lens, making a playful, funny face.
[Shot 28] At 00:13.500, a rapid zoom-out as she stands up and adjusts her oversized blazer.
[Shot 29] At 00:14.000, a handheld camera rapidly pulls backward. <Subject 1> stands under the bus stop sign, waving goodbye.
[Shot 30] At 00:14.500, a final radial blur contracts the image, snapping to black.
overall_soundscape:
Ambient city traffic noise, distant sirens, and the sound of wind are present but heavily chopped and mixed low to match the frantic cuts. Accentuated sound effects of footsteps, clothing rustles, and camera "whooshes" are synchronized with every whip-pan and crash zoom.
non_diegetic_music:
A hyper-fast, aggressive electronic or breakbeat track with a heavy, distorted bassline drives the editing. The beat drops and snare hits are perfectly synchronized with the whip pans, crash zooms, and radial blur transitions, amplifying the chaotic, high-energy visual rhythm. <lora:minimax_h3_fl2v_turbo_8step_v1.0_comfyui_bf16:1> <lora:minimax_h3_fl2v_turbo_8step_v1.0_comfyui_bf16:1>
A dark urban fantasy story in one continuous shot: a chase through a neon alley, a door spray-painted on a dead-end wall that becomes real, a field of giant flowers behind it, and the pursuers left in front of a flat painting. The 20-step clip keeps the whole story; a 12-step turbo version of the same prompt came out grainy and lost part of it.
| Model | Size | Frames | Sampler | Steps | Shift | Seed | Time |
|---|---|---|---|---|---|---|---|
| Standard | 576×1024 | 362 (15.08 s) | Res Multistep | 20 | 12 | 20261002 | 1628.1 s (27 min) |
Prompt
Single continuous cinematic shot with coherent character identity throughout the entire video.
A dark, stylish urban fantasy chase scene at night. A beautiful 25-year-old adult woman with a fashionable punk style runs through a narrow back alley in the city, chased by several strange and threatening figures. She has short layered black hair with subtle crimson and violet streaks, pale skin, sharp eyes, and a cool, fearless expression. She wears a fitted black leather jacket with metal studs and zippers, a dark top, slim black pants, heavy boots, layered silver accessories, and a small crossbody bag. Her look is fashionable, rebellious, elegant, and effortlessly cool.
The alley is moody and atmospheric, filled with concrete block walls, graffiti, pipes, fire escapes, dumpsters, puddles, drifting steam, and flickering neon signs. The wet pavement reflects red, violet, and cyan neon light. The lighting is low-key, contrasty, and cinematic, with glossy reflections and deep shadows. The air feels tense, dangerous, and stylish.
The camera moves dynamically in a sleek handheld cinematic style, tracking closely behind her, flowing down to low side angles and close views of her boots splashing through puddles, then smoothly arcing upward to her face and an over-the-shoulder view of the pursuers, all within one unbroken tracking shot. Her motion is fast, agile, and realistic. The people chasing her are unsettling and vaguely menacing, moving aggressively but still human, dressed in dark streetwear, partially obscured by shadow and steam.
She turns sharply into a dead-end alley and suddenly stops in front of a plain concrete block wall. For one tense second, she appears cornered. Her breathing is hard, but her expression stays controlled. She glances back and notices the pursuers hesitate for a split second as they enter the alley.
She calmly pulls a spray paint can from her bag. With precise, practiced movements, she rapidly paints the outline of a full-size door onto the bare concrete wall. The graffiti is elegant, clean, and visually striking, with smooth curves and sharp lines. Fresh paint gleams under the neon glow. As she completes the final stroke, the painted lines begin to emit a faint supernatural glow. The flat graffiti door slowly transforms into a real three-dimensional door embedded in the wall, while traces of wet spray paint remain visible around the edges.
She grabs the handle and pulls the door open.
Beyond the doorway is a completely different world: an enormous surreal flower field under a strange dark luminous sky. The flower field is vast and dreamlike, filled with gigantic flowers taller than a person, some towering like trees. Their petals are deep crimson, black-violet, midnight blue, and pale silver, with subtle glowing edges. The atmosphere is mysterious and elegant rather than cheerful. Mist hangs low over the ground. Tiny glowing particles drift in the air like pollen or ash. A cool wind moves through the giant flowers in slow waves. Soft shafts of otherworldly light break through the dark sky, creating a surreal contrast with the grim alley behind her.
She turns her head back one last time and gives the pursuers a small, confident, knowing smile. Then she steps through the doorway into the giant flower field and disappears into the other world.
The door swings shut behind her.
As the pursuers rush forward, the real door instantly collapses back into a flat spray-painted image on the wall. The alley is once again only a dead end. The strange pursuers stop abruptly, stunned and confused, staring at the painted door in disbelief.
Stylish dark fantasy, urban punk aesthetic, sleek cinematic pacing, realistic body motion, dramatic contrast between gritty neon alley and surreal flower world, seamless magical transformation, high detail, moody atmosphere, glossy wet textures, deep shadows, subtle film grain, shallow depth of field, high-end movie cinematography, visually coherent, elegant and cool.
Fifteen-second vertical 9:16 film. Keep the same woman, hairstyle, clothes and bag throughout. Narrative pacing: seconds 0-3, the neon alley chase; 3-4, she reaches the dead-end wall and glances back; 4-8, she retrieves the spray can and rapidly draws the full-size door; 8-10, the glowing painted door becomes solid and opens onto the giant dark flowers; 10-12, she gives her knowing smile and steps through; 12-15, the door closes, becomes flat paint again, and the pursuers stop in disbelief. The final image clearly shows the pursuers facing the flat painted door. Synchronized natural sound: boots splashing in puddles, urgent footsteps, controlled hard breathing, spray-can rattle and hiss, a subtle magical shimmering whoosh, cool wind rustling giant flower petals, and a final door slam. No dialogue, subtitles, text overlays or music.
Eight seconds, landscape, a cel-shaded style with a synthwave soundtrack and engine sound. The prompt describes the clip second by second.
| Model | Size | Frames | Sampler | Steps | Shift | Seed | Time |
|---|---|---|---|---|---|---|---|
| Standard | 1024×576 | 192 (8 s) | Euler | 12 | 12 | 479172522044947 | 380.6 s |
Prompt
0–1 second: A motorcycle rider speeds into view through a cel shaded cyberpunk city with strong ta86_inkriot energy and a synthwave sunset palette of neon pink, purple, orange and cyan. The camera tracks alongside the bike at a slight low angle, with the buildings and glowing signs already sliding smoothly through the background.
1–2 seconds: The rider accelerates down the street in a clean, controlled motion. The camera keeps pace while the environment moves with clear parallax, cel shaded buildings, power lines, glowing windows and streetlights drifting past without becoming chaotic.
2–3 seconds: The camera shifts slightly toward a three quarter rear angle, showing more of the road ahead. The sunset sky burns in warm orange and magenta tones while reflections and neon accents streak gently across the bike and the pavement.
3–4 seconds: The rider leans into a smooth turn through the city block. Background buildings, roadside barriers and signs move dynamically with the turn, giving a strong sense of speed while keeping the motion readable and stylish.
4–5 seconds: The bike comes out of the turn and charges down a longer stretch of road. The camera pulls back slightly, revealing more of the surrounding city in bold cel shaded shapes, with synthwave colors glowing across rooftops and windows.
5–6 seconds: A subtle camera drift to the side adds more motion as the rider passes another row of stylized buildings and neon storefronts. The sunset light mixes with the electric city glow, keeping the inkriot and synthwave mood strong.
6–7 seconds: The rider pushes forward at high speed, body steady and confident. The road markings, nearby walls and street objects slide by faster, while the far background continues moving in a smooth layered way.
7–8 seconds: The camera eases into a slightly wider tracking shot as the motorcycle continues racing into the glowing city. The scene ends with the rider still in motion, framed by the sunset sky and the cel shaded cyberpunk skyline.
Audio: A steady synthwave beat throughout, energetic but clean, paired with clearly audible motorcycle engine sound, tire noise on the street, light wind rush, and subtle city ambience. No dialogue.
These use img2img: the input picture is the first frame of the video. For pictures that define a person or an object rather than a frame, see Reference Pictures. The pictures are not from H3: they come from the test set of the Qwen-Image 2.1 extension, resized to the video size. See First and Last Frame for how to set it up.
| Input picture | Result |
|---|---|
 |
 |
| Model | Size | Frames | Sampler | Steps | Shift | Seed | Time |
|---|---|---|---|---|---|---|---|
| Standard | 768×1152 | 73 (3.04 s) | Res Multistep | 20 | 12 | 7 | 349.8 s |
Prompt
A young woman rides her bicycle along a sunlit park path toward the camera, her hair moving in the wind and the flowers in the basket swaying. The camera slowly moves back in front of her. Sound of the bicycle chain, birds singing, a light breeze and her soft laugh. No music, no subtitles.
| Input picture | Result |
|---|---|
 |
 |
| Model | Size | Frames | Sampler | Steps | Shift | Seed | Time |
|---|---|---|---|---|---|---|---|
| Standard | 1152×768 | 73 (3.04 s) | Res Multistep | 20 | 12 | 7 | 292.4 s |
Prompt
A young woman sits by the window of a train at night as rain streaks down the glass and city lights slide past outside. She slowly lifts her head and looks out of the window. Sound of the train on the rails, rain against the window and a muffled station announcement. No music, no subtitles.
Tip
Use the model's native size for pictures: 768 pixels on the short side (768×1152, 1152×768). The same pictures at 384×576 came out distorted, especially with the turbo LoRA.
The same two pictures with the turbo LoRA at 8 steps, in about half the time: Speed Comparisons.
A short line of dialogue with kitchen sounds around it: the knife on the board, the pan, the extractor fan.
| Model | Size | Frames | Sampler | Steps | Shift | Seed | Time |
|---|---|---|---|---|---|---|---|
| Standard | 1152×768 | 73 (3.04 s) | Res Multistep | 20 | 12 | 101 | - |
Prompt
A woman in her thirties cooks dinner in a small, warm home kitchen in the evening. She chops carrots on a wooden board, tastes the sauce from a wooden spoon, turns to someone off camera and says with a smile: "This needs a little more salt." Sound of the knife on the board, the pan sizzling and the extractor fan humming. Natural light from the window, handheld camera. No music, no subtitles.
| Model | Size | Frames | Sampler | Steps | Shift | Seed | Time |
|---|---|---|---|---|---|---|---|
| Standard | 1152×768 | 73 (3.04 s) | Res Multistep | 20 | 12 | 102 | - |
Prompt
Two women friends in their twenties sit at a small table in a busy neighborhood cafe. One leans forward, excited, and says: "You won't believe what happened today." The other laughs and covers her mouth. Sound of cups and saucers, the espresso machine and quiet chatter around them. Soft daylight, medium shot. No music, no subtitles.
An absurd sequence in order: water and clothes, then fish, then a swimming dog and a seal. A good test of ordering events in time.
| Model | Size | Frames | Sampler | Steps | Shift | Seed | Time |
|---|---|---|---|---|---|---|---|
| Standard | 448×672 | 158 (6.58 s) | Res Multistep | 20 | 12 | 20261003 | 196.9 s |
Prompt
A scene in a laundromat. One of the washing machines bursts open and an endless torrent gushes out of it. Along with the water come out clothes, then fish, then a swimming dog, and finally a seal.
One continuous shot with a steady wide view of the washing machine and the flooded laundromat floor. The absurd sequence is clearly visible in order: the machine bursts open, water and clothes gush out, then fish, then a dog swimming with its head above the water, and finally a seal. The seal is visible by 5.5 seconds; the torrent continues through the end. Realistic water splashes, wet textures, believable animal motion, cinematic lighting. Natural sound of the washing machine, the burst, rushing water and splashes. No dialogue, music, subtitles or text overlays.
A spice seller speaks Arabic over the noise of the souk, and a gust of wind lifts the red powder through the sunlight. Made with ck attention, about 7% faster per step than the default on this card (see ck attention).
| Model | Size | Frames | Sampler | Steps | Shift | Seed | Time |
|---|---|---|---|---|---|---|---|
| Standard + INT4 text encoder | 960×544 | 124 (5.17 s) | Res Multistep | 20 | 12 | 801 | 253.9 s |
Prompt
integrated_multimodal_description: [Shot 1] Live-action, cinematic, a medium shot frames a spice stall in a crowded souk in Marrakech in the late afternoon, pyramids of red paprika, yellow turmeric and green henna powder under hanging brass lanterns. The spice seller, a man in his fifties with a short white beard and a striped djellaba, with a deep, friendly voice (S1), scoops saffron threads into a small paper cone, holds it up to the camera and says: <d>[Arabic] هذا أحسن زعفران في السوق كله!</d> The camera pushes in with small amplitude at slow speed as he laughs and pours the saffron into a customer's open hands, and a gust of wind lifts a puff of red powder that drifts through the slanting sunlight.
overall_soundscape: Busy market chatter, a donkey cart rattling past on the cobbles, the rustle of the paper cone and the distant calls of other sellers.
non_diegetic_music: N/A
Made on a 16 GB card (RTX 2000 Ada, 31 GB of system RAM) with the GGUF Q4_K checkpoint and Never OOM on: a beekeeper speaks Greek over the hum of the bees. The setup is in 16 GB Cards.
| Model | Size | Frames | Sampler | Steps | Shift | Seed | Time |
|---|---|---|---|---|---|---|---|
| GGUF Q4_K + INT4 text encoder | 640×384 | 73 (3.04 s) | Res Multistep | 20 | 12 | 901 | 272.2 s |
Prompt
integrated_multimodal_description: [Shot 1] Live-action, cinematic, a medium close-up frames an old beekeeper in a white veil and gloves among wooden hives on a sunny hillside of wild thyme in Crete. He lifts a frame heavy with golden honeycomb out of the hive, bees swirling around it in the warm light, tilts it toward the camera and (S1) says proudly in a calm, gravelly voice: <d>[Greek] Κοίτα, καθαρό μέλι θυμαριού!</d> The camera pushes in with small amplitude at slow speed as a drop of honey slides down the comb.
overall_soundscape: A dense, steady hum of bees, cicadas in the dry grass and the wooden creak of the frame lifting out of the hive.
non_diegetic_music: N/A
A rotation in the air is hard for video models: the skater has to turn half around and land rolling away backwards. With Sparse Attention in its default window the skater turned and came back facing the camera; starting it at 50% of the steps kept the turn. This clip is without Sparse Attention.
| Model | Size | Frames | Sampler | Steps | Shift | Seed | Time |
|---|---|---|---|---|---|---|---|
| Standard | 1024×576 | 124 (5.17 s) | Res Multistep | 20 | 12 | 42 | 324.6 s |
Prompt
A young skateboarder rolls across a sunny concrete skatepark toward a low ramp, pops an ollie and spins a full 180 degrees in the air, landing backwards and rolling away from the camera. The camera tracks him at waist height. Sound of the urethane wheels on concrete, the snap of the tail, the landing clack and a light echo off the bowl. No music, no subtitles.
The FastH3 8-step V2 checkpoint, a distilled H3 by FastVideo, with the same text encoder and VAEs. Sampling took about 31 seconds (3.9 s per step); the whole request was not timed.
| Model | Size | Frames | Sampler | Steps | Shift | Seed | Time |
|---|---|---|---|---|---|---|---|
| FastH3 8-step V2 (INT8) | 960×544 | 73 (3.04 s) | Res Multistep | 8 | 10 | 11 | - |
Prompt
A middle-aged radio host sits in a warm, dimly lit studio, leans toward the microphone and says clearly: "Good evening, and welcome to the late show." He smiles. Close-up, soft studio light. Only his voice and the quiet hum of the room. No music, no subtitles.
The same prompt and seed with the standard model at 20 steps and with FastH3 at 8 steps with its own sparse attention (VSA). Both are good. FastH3 took a little more than half the time (the time includes switching checkpoints) and came out louder; the standard one sounds more natural for the place. The preview shows both side by side; the MP4 plays them one after another, each with its own sound. Each clip alone: standard, FastH3.
| Model | Size | Frames | Sampler | Steps | Shift | Seed | Time |
|---|---|---|---|---|---|---|---|
| Standard | 1152×768 | 73 (3.04 s) | Res Multistep | 20 | 12 | 5 | 257.1 s |
| FastH3 8-step V2 + VSA | 1152×768 | 73 (3.04 s) | Res Multistep | 8 | 10 | 5 | 146.9 s |
Prompt
An old fisherman with a white beard sits on a weathered wooden pier mending a fishing net. He looks up toward the camera, smiles and says in a gravelly voice: "The sea always gives back what it takes." Seagulls cry overhead, the wood creaks and small waves lap against the posts. Golden afternoon light, close-up. No music, no subtitles.
H3 Eros Max beta5 is a community fine-tune of H3 published on Civitai, with the turbo distillation already merged in: it is meant for 8 steps without a LoRA. It is an adult-oriented model; it was tested here only with ordinary scenes, and these clips are all it shows on this page. Two of its files load in Forge Neo as they are, with the standard text encoder and VAEs: the INT8 file (Star Converter format) and the 13 GiB file labelled fp8, which is actually a W4A8 format. See Models.
All five use the INT8 file, Res Multistep, 8 steps, Shift 12. The first four are 1152×768 with 73 frames and took about 140 s each; the prompts are the kitchen and cafe ones above and two more below.
| Kitchen (seed 101) | Cafe (seed 102) |
|---|---|
 |
 |
| Morning jog (seed 103) | Delivery (seed 104) |
 |
 |
Prompts: morning jog and delivery
Early morning in a city park. A man in his forties jogs along a gravel path toward the camera, breathing steadily, wearing a gray hoodie. Mist over the grass, low sun behind the trees. Sound of his footsteps on the gravel, his breathing, birds and a dog barking in the distance. Tracking shot. No music, no subtitles.
A delivery driver in a brown uniform walks up to the front door of a suburban house holding a cardboard box, rings the doorbell, smiles when the door opens and says: "Package for you!" Sound of the doorbell, a car passing on the street and birds. Afternoon light, eye-level shot. No music, no subtitles.
The bus stop prompt without a LoRA: 576×1024, 362 frames, 8 steps, Shift 12, same seed, 777 s (13 min). The RAM peak was the same 46.3 GiB.
Clips the author generated in the Forge Neo WebUI on the test machine, with the author's own prompts, kept as Forge saved them. FastH3 and H3 Eros Max were run here at 20 steps instead of their usual 8.
| 5 seconds | 9.4 seconds |
|---|---|
 |
 |
| 576×768, 124 frames, seed 2146798990 | 640×832, 226 frames, seed 3088296760 |
FastH3 8-step V2, Res Multistep, Simple, 20 steps, CFG 1, Shift 12.
Prompt
a quarter view of a full body shot of a cute Japanese woman of her 21s, she is wearing a pink tight puff short sleeve Jiraikei shirt, the shirt is decorated with rich and varied black frills, black pleated layered ultra-mini suspender skirt decorated with pink frill borders, black over knee socks, black platform shoes, fancy cross body bag, she has detailed face, beautiful eyes, detail eyes, long eyelashes, double eyelids, beautiful pink lips, heavy makeup, porcelain skin, short and thick legs, black long hair, earrings, necklace, skinny body, large , she is standing back against a low stone wall at the edge of a crowded sidewalk surrounding a park in the bustling heart of Tokyo at night, she is watching her mobile with a gloomy face, the background consists of a stone wall stretching behind her and a row of 5 women in various cute fashions standing at the same intervals along the wall next to her while watching their mobiles, moody lighting, cinematic lighting, cinematic composition, masterpiece, photo realistic,8k,ultra-detailed, quarter view from above
One continuous tracking shot where only the woman and her dress keep their colour. H3 Eros Max beta5 (INT8), 576×832, 209 frames (8.7 s), Res Multistep, Simple, 20 steps, CFG 1, Shift 12, seed 2818957432.
Prompt
A beautiful woman walks confidently toward the camera on a busy city sidewalk during a bright sunny day. She wears stylish sunglasses and a flowing vivid red dress that moves naturally with every step.
She is the only person in color.
Everyone else around her, men and women walking in both directions, wears formal black suits and appears completely monochrome, black and white, including their skin, clothes and accessories. The surrounding city environment is also black and white.
The woman remains vivid and colorful, with the red dress strongly contrasting against the monochrome world.
The camera smoothly tracks backward in front of her at walking speed. Natural body movement, realistic fabric physics, subtle wind moving her dress and hair, pedestrians naturally passing beside her, realistic sunlight and shadows, cinematic shallow depth of field, high contrast, photorealistic.
No slow motion. No camera cuts. One continuous shot.
Two K-pop singers introduce their group in Korean over a trap beat, in one continuous shot that slides from one to the other. The clip starts on a picture of the first singer and ends on a picture of the second (first and last frame). What made it work: the music is described before the scenes and starts "at the very first second", and the prompt asks for one shot with no cuts and both singers in the frame during the move. Standard model with the INT8 text encoder, 576×768, 362 frames (15.08 s), Res Multistep, 20 steps, Shift 12, seed 2026, 1086.5 s (18 min).
Prompt
One continuous shot with no cuts, on a single stage at night: two South Korean female pop artists of the group Digital Queens stand side by side in front of the same warm city lights blurred into bokeh, under the same soft cinematic stage lighting, ultra realistic.
A hard-hitting K-pop trap beat at 160 BPM with a heavy, booming 808 bass and crisp hi-hats starts at the very first second and plays continuously for the whole 15 seconds, mixed just below the voices so every word stays clear.
0-6 seconds: Close on the first artist, long straight black hair, sparkling pink sequined dress and a small silver necklace, elegant and warm. She looks into the camera, smiles and introduces the group and herself in Korean in a soft, polite voice: "안녕하세요, 저희는 디지털 퀸즈예요! 저는 김서연입니다." She makes a small finger-heart gesture near her face.
6-8 seconds: Without cutting, the camera slides sideways along the same stage; for a moment both artists are in the frame together, side by side under the same lights, then the camera settles on the second artist.
8-15 seconds: The second artist, short messy platinum blonde hair, smoky eye makeup, black leather jacket off the shoulders, straps, a studded choker and silver chains, rebellious and defiant, looks up at the camera with a smirk and says in Korean in a confident, casual voice: "그리고 난 박민지. 디지털 퀸즈, 잘 부탁해!" She ends with a playful defiant gesture and sticks out her tongue.
No subtitles, no text.
| Page | What it shows |
|---|---|
| Reference Pictures | Ref2VA with 1, 3 and 9 reference pictures, with prompts in Italian, German and French |
| First and Last Frame | A clip that starts and ends on given pictures |
| Format Comparisons | The same scenes with INT8, W4A8, INT4BQ, GGUF and the INT4 text encoder, on an A40 and an RTX 4090 |
| Speed Comparisons | Anime, claymation and wildlife scenes with ck attention, Sparse Attention and the turbo LoRA with two samplers; turbo clips from a picture |
| Getting Started | The first test: a bird |
| 16 GB Cards | A wizard, a tea seller and a street cook made on a 16 GB card |
| Bloopers | Clips that went wrong, why, and how to avoid it |
| All Generations | Every clip made while testing, variants included, with its prompt |
| Clip | Model | Size | Frames | Steps | Time |
|---|---|---|---|---|---|
| Bus stop | Standard + turbo LoRA | 576×1024 | 362 | 8 | 840.8 s |
| Neon door | Standard | 576×1024 | 362 | 20 | 1628.1 s |
| Motorcycle | Standard | 1024×576 | 192 | 12 | 380.6 s |
| Bicycle, first frame | Standard | 768×1152 | 73 | 20 | 349.8 s |
| Night train, first frame | Standard | 1152×768 | 73 | 20 | 292.4 s |
| Kitchen | Standard | 1152×768 | 73 | 20 | - |
| Cafe | Standard | 1152×768 | 73 | 20 | - |
| Laundromat | Standard | 448×672 | 158 | 20 | 196.9 s |
| Marrakech | Standard + INT4 text encoder | 960×544 | 124 | 20 | 253.9 s |
| Beekeeper | GGUF Q4_K, 16 GB card | 640×384 | 73 | 20 | 272.2 s |
| Skatepark 180 | Standard | 1024×576 | 124 | 20 | 324.6 s |
| Radio host | FastH3 | 960×544 | 73 | 8 | - |
| Fisherman | Standard / FastH3 | 1152×768 | 73 | 20 / 8 | 257.1 s / 146.9 s |
| Eros Max, four scenes | Eros Max INT8 | 1152×768 | 73 | 8 | about 140 s each |
| Eros Max, bus stop | Eros Max INT8 | 576×1024 | 362 | 8 | 777 s |
| Jiraikei, Tokyo | FastH3 | 576×768, 640×832 | 124, 226 | 20 | - |
| Red dress | Eros Max INT8 | 576×832 | 209 | 20 | - |
| Digital Queens | Standard, first + last frame | 576×768 | 362 | 20 | 1086.5 s |
More measurements, including memory, are in Performance and Memory.
Commercial use of H3 outputs needs a commercial license from MiniMax; check the model's license before using clips like these.