Experiment with structured poster prompting in MOSS-Image SFT. The connected positive text encoder contains a JSON-like layout specification with a canvas summary, labeled text and object items, bounding boxes, and coordinate format. The workflow uses that text as a prompt; there is no separate layout-control node.
Main features:
- MOSS-Image SFT INT8 ConvRot model, Qwen3.5 9B BF16 text encoder, and FLUX.2 VAE.
- A single output-reachable Euler sampler at 50 steps, CFG 3, simple scheduler, fixed seed 8001.
- Example split-panel travel poster with the word EXPLORE and a blue mountain illustration; SaveImage produces the final image.
Suggested workflow:
1. Install the required model files and load the JSON in ComfyUI.
2. Edit the positive prompt's summary, items, text, and bounding boxes while retaining valid JSON syntax.
3. Generate and inspect the final layout and typography. Iterate on coordinates and wording where placement differs from the prompt.
Required model filenames: moss_image_sft_int8_convrot.safetensors; qwen3.5_9b_bf16.safetensors; flux2-vae.safetensors. Text and bounding-box adherence are model-dependent.
RunningHub AI workflow: https://www.runninghub.ai/zh-cn/post/2113906852581289985?inviteCode=rh-v1111
RunningHub signup: https://www.runninghub.ai/?inviteCode=rh-v1111
Bilibili video: https://www.bilibili.com/video/BV1ZHHP6FEVZ/
Related model files: https://pan.quark.cn/s/20e19307e8ef