Loop ForgeSwap text: Kate's page
Works, from MOSS-Image add-ons, tested locally
- Model
- MOSS-Image Base (moss_image_base_int8_convrot.safetensors)
- Output
- 880×1184
- Sampling
- 50 steps, CFG 3, er_sde / normal
- Seed
- 0
- Without LoRA
- 184.3 s, peak VRAM 10.1 GiB
- Swap text
- 102.6 s, peak VRAM 10.5 GiB
- Ran on
- RTX 3060, 12GB (local)
What was wired in
Input page
Kate's comic page (swap text)Comic page drawn from the Kate cast photo with the Qwen-Image 2.1 edit template (not MOSS)Download PNG11.4 MB
On all three pages it lettered the line cleanly and left the rest of the page alone. Without the LoRA the page comes back the same.
"summary":"caption":"comic, english text","items":["label":"text","caption":"text","text":"REALLY?","bbox":[-500,255,-190,410]],"bbox_coord_space":"abs_centralized","bbox_format":"yxyx"
Notes
Swap text is its own LoRA (moss_i2i_swap_text) on MOSS Base, loaded by the MOSS i2i Reference node together with the input page. It does not find or translate the bubble itself: the English line and the bubble's box are typed into the prompt as layout JSON.
Times on this page are for these two runs.
sorryhyun/MOSS-I2I-LORAs on Hugging Face
Run it yourself
Download everything for this test (.zip, 11.4 MB). It holds the workflow, every input it loads and a README. Copy the inputs into ComfyUI/input, put the models below in place and drag the workflow onto ComfyUI.
| File | Put it in | Size |
|---|---|---|
| moss_image_base_int8_convrot.safetensors | models/diffusion_models | 5.4 GB |
| qwen3.5_9b_w4a8_gptq.safetensors | models/text_encoders | 7.3 GB |
| flux2-vae_small.safetensors | models/vae | 249.5 MB |
| moss_i2i_swap_text.safetensors | models/loras | 149.4 MB |
Custom nodes: ComfyUI-MOSS-Toolkit at commit 53ee3b3. Everything else is part of ComfyUI 0.43.1 (frontend 1.57.6, templates 0.11.100).