Fibo 2.0 in ComfyUI: GGUF and W4A8 workflow setup

This walkthrough covers Fibo 2.0 image generation and editing in ComfyUI, with GGUF and W4A8 downloads aimed at lower VRAM use. The presenter provides Q4 through Q8 quantizations and reports model files ranging from 6.5 to 10 GB. Those figures describe download sizes; the walkthrough does not establish a minimum VRAM requirement. He also says users need Bria's license approval before using the model.

The setup connects an input image and prompt to Fibo 2 text encoding, with an optional Qwen3-VL 8B W4A8 prompt enhancer. Other components include a Qwen3-VL 4B int8 text encoder and the Flux 2 VAE. The recommended sampler settings are four steps with CFG 1 for turbo editing, or 30 steps with CFG 5 for the dev model's generation and editing tasks.

Generation examples cover portraits, product images and posters with embedded text. Turbo editing tests include anime and 8-bit style transfers, clothing changes and object insertion. The presenter reports several failures: an added person looks pasted in, inserted products have incorrect scale, and text from a reference image carries into the result. He recommends the dev model at full step count for more polished multi-reference edits. A removal test leaves clothing behind until the prompt explicitly asks to remove it, while a thumbnail layout improves after a resolution change.