FIBO 2.0 Generate + Edit | 4-Step Turbo, Auto Prompt

Say what to change in plain words, and FIBO 2.0 edits your picture in 4 steps. Switches write the prompt for you, make a new picture instead, or make the result twice as big with SeedVR2.

Want Qwen Image 2.1 instead? My Qwen Image 2.1 Generate + Edit and its Turbo version work the same way.

How to use it

1. Load your picture.

2. Write the change, like "make it snow". With Auto prompt on, a few words are enough.

3. Run. A before and after slider shows your picture against the edit.

4. To make a new picture instead, switch Edit a picture off, write a prompt and set the size in Latent Size.

The switches

• Edit a picture. On as the file opens: the prompt changes the picture you load, at about 1 megapixel in your picture's own shape. Off: the prompt makes a new picture.

• Auto prompt. FIBO reads a long, detailed prompt. Qwen3-VL 8B writes it from a few words of yours, and for an edit it looks at your picture first. What it wrote shows in Show Text. Off, your words go to FIBO as you typed them.

• Upscale the result. SeedVR2 makes the result twice as big on each side and sharpens it.

What you need

• ComfyUI's latest development version. FIBO 2.0 support went into ComfyUI on 2026-10-21 and is not in a release yet. Update ComfyUI from GitHub (Windows portable: update\update_comfyui.bat). The desktop app gets it with ComfyUI's next release.

• AusBoss nodes 2.13.0 or newer. In ComfyUI-Manager, search "AusBoss". Already have them? Update first.

• The models:

  • fibo2_edit_turbo_int8_convrot.safetensors → models/diffusion_models · 8.41 GB

  • qwen3vl_4b_int8_convrot.safetensors → models/text_encoders · 4.86 GB

  • flux2-vae.safetensors → models/vae · 336 MB

• Have FLUX 3 Image Klein already? The text encoder and the VAE are the same files, so you only need the 8.41 GB model.

• For Auto prompt, Qwen3-VL 8B: qwen3vl_8b_int8_convrot.safetensors → models/text_encoders · 9.35 GB (the same file my Qwen Image 2.1 pages use)

• Optional, uncensored versions of both text encoders (community-made "Heretic" files; the regular ones stay picked, see "Uncensored text encoders (optional)" in the workflow's note) → models/text_encoders:

  • qwen3-vl-4b-heretic_int8_dynamic_convrot.safetensors · 4.84 GB, for Text encoder

  • qwen3-vl-8b-heretic-1.3.0_fp8_e4m3fn.safetensors · 10.02 GB, for Text encoder for Auto prompt

• Only for Upscale the result, SeedVR2:

  • seedvr2_3b_int8_convrot.safetensors → models/diffusion_models · 3.46 GB

  • seedvr2_ema_vae_fp16.safetensors → models/vae · 501 MB

Tips

• Say what changes and what stays: "Change her black t-shirt to a red sweater. Keep everything else the same."

• Keep changing the same picture without it getting crunchy: pick Erase on the before and after slider, paint where the edit should show (the rest stays your picture), press Save, drop the saved picture on the picture node and run the next edit. Copy, a click on the picture node and Ctrl+V work too (Copy needs ComfyUI opened as localhost or https).

• It barely moves your picture. In my tests a typical edit moved it about a third of a pixel, so the slider lines up.

• Don't like the result? Press New seed on the Seed node. With Auto prompt on, you can also fix the wrong part in Show Text, copy it into your prompt and run with Auto prompt off.

• It doesn't know famous people or game characters by name. Describe how they look instead.

• "Zoom out" doesn't zoom out. It changes what's in the picture, not the framing.

• Times on an RTX 5090: an edit takes about 3 seconds, a new picture about 2. Auto prompt adds about 6 to 14 seconds, depending on how long a prompt it writes, and Upscale the result about 3 to 4.

• LoRAs made for FIBO 2.0 go in LoRA Loader.

What it does not do

• It takes one picture. For an outfit from another photo, see my Qwen Image 2.1 Turbo Multi-Image Edit page.

• It is for non-commercial use. See Credits.

Credits

FIBO 2.0 is by Bria AI, under CC BY-NC 4.0: non-commercial use only. For commercial use you need a license from Bria. The ComfyUI files are from Comfy-Org. SeedVR2 is by ByteDance Seed. Qwen3-VL is by Alibaba Qwen.

v1.0 (2026-10-21): first release.

Questions or bugs: GitHub issues or the comments here. Videos on YouTube @ausbossdev. More from me on X @Zanzibased and GitHub.