What works where today
| Task | Hosted studio | Local ComfyUI |
|---|---|---|
| Generate a new image from text | Available | Available |
| Guide text and objects with JSON regions | Available | Available |
| Change the style of an existing image | Not available yet | Community Edit LoRA |
| Repaint part of an image | Not available yet | Edit workflow with a mask or dedicated inpainting workflow |
| Replace words inside an image | Not available yet | Edit LoRA, followed by careful proofreading |
| Preserve a composition while changing its contents | Not available yet | Reference-image editing or a reconstructed JSON layout |
Which workflow would you use first?
Select every option that applies. Your choices determine what we evaluate first.
We do not accept image uploads on this page. We record your selected workflows and optional note. If you provide an email, it is encrypted for a possible early-access invitation and is not added to a marketing list.
Run image editing locally today
The current community workflow uses a MOSS-Image Edit LoRA with a compatible model and reference-image conditioning in ComfyUI. For native loaders, keep the diffusion model in models/diffusion_models/, Qwen3.5 encoder in models/text_encoders/, VAE in models/vae/, and Edit patch in models/model_patches/.
Do not mix the two loading paths: a patch loaded with the native model-patch loader belongs in models/model_patches/; a LoRA weight loaded with a LoRA loader belongs in models/loras/.
Use a workflow written for the exact Edit patch and loader path. A generic text-to-image LoRA graph does not provide the reference-image conditioning required for editing.