Revision Log

0.6.3 (2026-10-09)

Pictures carry labels, @img1, @img2, and so on, in the forms, in the chat, and on the cards. Use them in a description to say which picture is which; the model receives "image 1", "image 2".
An arrow on a picture moves it earlier. A label with no picture stops the request before anything is sent.
Cards show the pictures a job used, with their labels, and the wording the model received.

Settings that cannot apply together no longer show together. For Seedream 4.5, Width and Height appear only when Size is custom, and Shape only when it is not; FLUX.2 dev and FLUX 1.1 pro show Width and Height only for a custom shape. Settings that do not apply are never sent.

A job with several pictures shows all of them: a row of small pictures under the large one, and arrow buttons in the full-size view.
Save no longer copies a job a second time when it is already in your folder (for example, after automatic saving). The button reads Saved and shows where the copies are.
Right-click a picture on a card for Save as, Show in folder, and Copy picture.
Edit, Expand, Upscale, Make a video, and Use as reference work on the picture shown large, not always the first one.

An alternate LoRA trainer: Qwen Image, on Replicate's qwen/qwen-image-lora-trainer, chosen under Trainer on the LoRAs tab.
Qwen LoRAs learn from plain words such as "a person named Pao" in place of a made-up trigger word.
A new picture model, Qwen Image with a LoRA, the only Qwen model that can run these LoRAs. It runs your trained model's own version.
Every LoRA shows the model it was made for, and Saved LoRA lists only the LoRAs that fit the chosen model.

LoRA training shows its stage: starting, writing captions for the photos, loading the FLUX model, training, and saving.
Before the steps begin, the bar moves back and forth so a training never looks stuck. During the steps it shows about how many minutes are left, and each training shows when it was last checked.

A shape typed in a chat message, such as 4:3, now reaches the model's shape setting. Before, it stayed in the prompt and the picture came out square.
Any setting named in a chat message goes into the request: the number of pictures, length, sound, resolution, safety, seed, steps, guidance, file type, style, what to keep out, and any other setting by its name on the model's card.
The plan lists every setting it will send, names the settings the model does not have, and says when a value stays at the default.
OpenRouter video plans show a price worked out from OpenRouter's own price per second.
Upscale in the chat takes "2x" and "faces"; Expand takes "safety on".

The chat model can be a private model in Ollama on this PC or a model on OpenRouter, and you can switch between them at any time. A badge shows which: a crossed-out eye for private, a cloud for OpenRouter.
Settings, under Chat, adds more OpenRouter chat models; OT Studio checks that each one can answer in a fixed format.
The Chat tab now has a left panel, like Pictures and Videos, with the chat model, the picture, edit, and video models the chat uses, and the Clear buttons.

A Chat tab: ask for pictures, edits, new shapes, upscales, videos, and longer videos in plain words. A model in Ollama on your PC reads the messages, and your chat text stays on your computer.
Every request shows what will happen, the picture it uses, and the price, and starts only when you click Go.
Chat results also get their full card on the Pictures or Videos tab.
Right-click a chat result for Save as, Show in folder, and Copy picture.
Every finished picture and video can now be saved to your folders automatically (on at first).
Clear chat keeps the pictures. Clear chat and delete its pictures moves the chat's saved copies to the Recycle Bin.

A LoRAs tab: your FLUX dev LoRAs with Use for pictures, Strength, Export, and Remove.
Import LoRA files from the Android app and the LoRA Trainer, or add a LoRA by its reference.
Train a LoRA on Replicate from the app, with photo checks, upright JPEG copies, progress, and Stop. The finished LoRA joins your list.
LoRAs trained into your private Replicate model run on their own version, because the public LoRA model cannot read private models.
A Saved LoRA chooser on the Pictures tab for FLUX dev with a LoRA.

Edit: opens a picture with the editing model you used last, ready for a description of the change and more reference pictures.
Expand: gives a picture a new shape with FLUX Fill dev, then rebuilds it at full size with the original pixels kept.
Upscale: 2 or 4 times with Real-ESRGAN. Pictures that are already large get a plain explanation instead.
Make it longer: the last frame of a video starts the next part, and OT Studio joins the parts into one video with its bundled ffmpeg.
Repeat on these jobs opens the same tool again with the same choices.

First version of OT Studio for Windows, with Pictures, Videos, and Settings.
Replicate, SpicyAPI (text only), and OpenRouter pictures and videos.
OpenRouter shows each model's own resolutions, shapes, lengths, frames, and sound switch.
Add any picture or video service by pasting its curl, Python, or Node example. Services can be exported and imported without keys.
History with Save, Open folder, Repeat, Delete, and Clear history.
Keys encrypted by Windows. Network access limited to approved hosts.