1.2.0

The workspace becomes the whole studio, and the studio grows a camera and a set
of lights. Everything below is in this release; the sections are grouped by what
they change rather than by the order they were written.

The Camera Studio

• A top-view stage inside the Prompt Frame and on its own Camera tab: place
  people, objects, walls, doors and windows, point a camera at them, and the
  geometry is written as the physical-camera language Krea 2 obeys - position,
  pitch in degrees, what the subject shows the lens, what is behind them, the
  lens itself. A bare "high angle" gets reinterpreted; the stack does not
• Lock on a subject or aim freely; walls occlude what the lens cannot see; a
  centred foreground person is named as between the camera and the subject
• Auto latent: the frame's aspect follows the camera, at a pixel budget
• Sets: built-in rooms (kitchen, living room, bedroom, bathroom, apartment) and
  two-person scenes (conversations, confrontation, boxing, baseball), plus your
  own saved and deleted from the panel. Every built-in set ships a light rig
• Camera paths: A to B in N shots, or an orbit sweep. The workspace renders one
  image per shot in the built-in sampler, each with its own camera words and its
  own camera-LoRA strengths, and the images come back as a batch
• Scene focus: the stage fills the screen with the subjects card beside it, Esc
  closes. Drag keeps the grab point; the wheel over a thing turns it
• Aperture: with an f-number set, the depth of field is computed from lens,
  aperture and distance, and says who is sharp and who dissolves
• Camera LoRAs card: zoom, height, orbit and back, each off / auto / manual,
  auto strengths coming from the same geometry the words come from. Three of the
  four sliders are ours and are a separate download; the zoom slider is
  Loraholic's and only ever linked, never shipped
• RedNode Camera Multi-Angle turns the same stage into a Qwen-Image-Edit
  Multiple-Angles prompt, for re-shooting an existing photo from a new viewpoint
• A master on/off for the whole Camera tab: off, nothing on it reaches a render

Lights

• Place lights on the stage like subjects: a ring sized by the real diameter,
  tinted by its colour, with a dashed line to whatever it is aimed at
• The sentence they write names a source, a direction and what it lands on,
  which is what the corpus study found gets obeyed. Hardness is real angular
  size, the key-to-fill ratio is inverse square, and named setups (Rembrandt,
  butterfly, split, rim, uplight, short, silhouette) come out of the geometry
• Exposure per light in stops, centred on 0, and colour from cold through
  neutral to warm, spaced in mireds. Moving a light dims it, as it should
• One-click rigs: three-point, Rembrandt, window, backlit
• Two optional lighting LoRA rows, off / auto / manual, driven by the rig you
  placed: brightness from the level your lights make, colour from the key
  light's temperature. Both files are third-party and only ever linked
• Style "Photo: nocturnal low-key" for the dark, velvety look that words alone
  fight for

Swaps, sets and passes

• Img2Img SWAP stage: put the Subject's face, head or whole person onto the
  picture with Alissonerdx's BFS LoRAs on Qwen-Image-Edit, before the Krea 2
  pass finishes it. Face, head and person modes, the Picture order read off the
  file name
• The same author's Krea 2 BFS files run instead as a Detailer pass, on two new
  per-pass options: Picture (the frame, or a detailer's crop, as the base
  reference) and a per-pass LoRA that never touches the main render. Presets for
  both head and body swap
• Img2Img RE-ANGLE: the source re-shot from another viewpoint by
  Qwen-Image-Edit-2511 and the Multiple-Angles LoRA before the i2i pass, from
  the bands or from the Camera tab's own studio
• LoRA sets: the LoRAs tab is Main plus any number of named sets, each on its
  own tab, and a rig, a Detailer pass or the paint pass picks one by name
• Detailer: a working resolution on sampler passes as well as detailer ones

NovelAI

• The NovelAI rig is public: a Models-tab rig kind that renders through your own
  NovelAI subscription instead of a checkpoint, with the site's own option set,
  vibe transfer and character prompts
• Vibe encoding is a button that says its price: 2 Anlas the first time for a
  picture, model and Information, then cached and free. Nothing spends Anlas on
  its own
• Your token lives in NAI_ACCESS_TOKEN or in the ComfyUI user directory, never
  in a workflow and never in the pack folder

Fixes worth naming

• An experimental reference toggle that rewrites the reference path now
  announces itself in the console: ref_t0_modulation cost a day of hunting a
  doubled, ghosted identity that no LoRA and no model was causing
• The Img2Img tab lights its dot on a wired canvas, prompt boxes fill their
  pane, the LoRA stack's footer sits under the stack instead of the node's
  bottom edge, and the select-mode bar appears where the rows are

1.1.0

Four new nodes. Video gets the filing system the pictures already had, and prompting gets
a node that treats word order as the setting it actually is.

• RedNode Save Video files a sequence into the same tree RedNode Save uses: the same base folder, subfolder, name and numbering tokens, the same drafts and keepers split, and the same readable text record beside the file. mp4, webm, gif or animated webp
• Save Video takes an audio track, and the prompt that made the clip. Sound shorter than the picture either plays once or repeats to the end, which is the choice ping pong forces on it
• Encoding looks for the ffmpeg that imageio_ffmpeg brings with it, then ffmpeg on PATH, then Pillow for gif and webp. A machine with none of those files the frames as an image sequence rather than losing a render to a missing codec
• RedNode Video Review plays what the graph just made, in the node, with a scrubber, a loop toggle and the previous runs in a strip. It files nothing, so leave it wired while you iterate. Feed it the path output of Save Video and it plays the clip that was really written instead of encoding a lookalike of it
• RedNode Prompt Frame puts Subject and Surroundings in their own boxes and orders them by framing, five steps from Portrait to Roomscale. Your wording is never rewritten, only the order and the joining words. Style blocks, twenty lighting setups, a brightness ramp and placement sit around that, and a second output calls out what quietly costs a render: a self-lit style fighting the lighting dropdown, a wide framing with nowhere for the subject to stand, a prompt run past the 90 to 150 word working range
• RedNode Describe To Boxes reads a picture into Subject, Surroundings and Light and colour on a local Ollama vision model, ready to wire straight into the Frame. The vision model is released from VRAM as soon as the reply lands, so it does not sit on top of the checkpoint for the rest of the queue
• Paint tab: a Passes box on the denoise row runs that same low denoise over its own result up to ten times inside one Generate, with a fresh seed each pass and the mask reapplied between them. The settling chain without the dragging, and only the last picture comes back
• Prompt Box values stop moving when the workflow changes. The panel was taking a slot in the saved widget list, so the prompt loaded into the seed, "fixed" into the wildcard toggle and the font size into the colour. Values are written and read by name now, and a workflow saved while it was wrong repairs itself on load
• Prompt Frame had the same fault and worse: its panel is built a frame late, so every load shifted Subject into Surroundings. Same fix
• Prompt Frame: a Push option under the Framing slider, off by default, for when a long prompt talks over the framing. Restate at the end repeats the shot as a closing sentence so the framing holds the end of the prompt as well as the front; Camera words swaps "small in the distance" for the label a caption would use, like "extreme long shot"
• Prompt Frame: the light conflict warning now also reads style text arriving on the wire, which is where a moodboard block lands and exactly where it never looked
• Prompt Frame: a tools fold like the Prompt Box has, so the seed, wildcard, font and colour rows tuck away under the panel. The text boxes also come back at whatever height you dragged them to
• Grabber: click the channel field and type to search, the same picker the LoRA rows use. Channels say how many values they carry, recently used ones come first, and a name that matches nothing is offered as a new channel
• The channel picker no longer filters by the name it already shows, which made the list a single entry of the thing you already had
• The example workflow ships as Rednode Ultima V1.1.0
• The README covers the prompt frame, and the node count on it is right again

1.0.36

Speed and polish, plus the first documentation pass.

• Thumbnails are served resized instead of full size, in the Review strip, the Save grid and the Workspace galleries. Large galleries and big pictures load far lighter
• Save browser: clicking a card no longer flashes blank or jumps the view
• README: nine screenshots, so the panel can be seen before installing
• The example workflow ships as Rednode_Ultima_V1.0.36 Release
• Use as reference: the renderer choice is repaired before the buttons read it, so they cannot briefly look enabled after a renderer is deleted and recreated
• Patch notes now ride every release automatically, which is why this list exists

1.0.35

Bug fix patch for the Masks redesign and the Paint tab.

• Masks: undo and redo now save the mask, so a queue renders exactly what the canvas shows
• Masks: changing the Paint on picture drops the previous picture's mask instead of carrying it over
• Masks: Open image works now, and a picture of your own survives a reload
• Masks: an auto mask base coat survives a panel rebuild instead of silently vanishing on the next save
• Paint Out: the Never shrink setting now works on external renderer chains, as it already did on Paint Render
• Whole frame: the Mask size dial works in both directions. A 3K picture with the dial at 1024 renders at 1024 and comes back at 1024, so a big frame can be painted at a working size and upscaled after
• Overlay colour and opacity changes repaint open Masks painters immediately
• Rendered by: a chain named Krea2 Workspace shows a warning that the main prompt renders and the paint prompt boxes are ignored
• Use as reference greys out on renderer chains that cannot carry references, with the reason on hover
• The registry's Documentation link no longer points at a folder that does not exist

The attached Rednode_Ultima_V1.0.35.json is the matching release workflow, cleaned of session state.