Fibo 2.0 GGUF

Fibo 2.0 is an 8B flow-matching image model with FLUX.2 VAE latents, a Qwen3-VL 4B encoder, a two-layer Perceiver, and a 44-block single-stream DiT. The Dev variant supports text-to-image generation and image editing. Edit Turbo is distilled for image editing.

Hugging Face repositoryfibo-2.0-GGUF on Hugging Face

Examples

The Dev workflow uses a structured prompt to make a product photo of a blue ceramic mug and a red apple.

The Edit Turbo workflow uses the prompt “Replace the blue ceramic mug with a red glass vase, preserving the apple, composition, lighting, and background.”

Available Quantizations

Dev

Quant	Size	File
Q8_0	8.29 GiB	fibo2-dev-Q8_0.gguf
Q6_K	6.48 GiB	fibo2-dev-Q6_K.gguf
Q4_K_M	5.16 GiB	fibo2-dev-Q4_K_M.gguf
Q8_CR	7.83 GiB	fibo2-dev_Q8_CR.gguf

Dev GGUF, Q8_0, 8.29 GiBfibo2-dev-Q8_0.ggufDev GGUF, Q6_K, 6.48 GiBfibo2-dev-Q6_K.ggufDev GGUF, Q4_K_M, 5.16 GiBfibo2-dev-Q4_K_M.ggufDev GGUF, Q8_CR, 7.83 GiBfibo2-dev_Q8_CR.gguf

Edit Turbo

Quant	Size	File
Q8_0	8.29 GiB	fibo2-edit-turbo-Q8_0.gguf
Q6_K	6.48 GiB	fibo2-edit-turbo-Q6_K.gguf
Q4_K_M	5.16 GiB	fibo2-edit-turbo-Q4_K_M.gguf
Q8_CR	7.83 GiB	fibo2-edit-turbo_Q8_CR.gguf

Edit Turbo GGUF, Q8_0, 8.29 GiBfibo2-edit-turbo-Q8_0.ggufEdit Turbo GGUF, Q6_K, 6.48 GiBfibo2-edit-turbo-Q6_K.ggufEdit Turbo GGUF, Q4_K_M, 5.16 GiBfibo2-edit-turbo-Q4_K_M.ggufEdit Turbo GGUF, Q8_CR, 7.83 GiBfibo2-edit-turbo_Q8_CR.gguf

Q8_CR uses the custom ConvRot layout and requires comfyui-gguf-reboot.

Model Input and Output

Variants and inputs

Variant	Inputs
Dev	Text prompt for image generation, or an image and editing prompt. The example workflow uses 50 steps and CFG 5.
Edit Turbo	Source image and editing prompt. This variant is for image editing, with 4 steps and CFG 1.

Outputs

Both variants produce images. Dev supports text-to-image generation and image editing. Edit Turbo is intended for image editing only.

ComfyUI Setup

Use ComfyUI core at commit 87ed5a066d9c4660ef42100fd01a888da6994856 or newer, with the comfyui-gguf-reboot loader nodes. Use UnetLoaderGGUFDynamicVRAM for the diffusion model. The loader nodes give the best performance.

Dependencies

Place the chosen GGUF file in models/diffusion_models/. Place these companion files in the folders shown.

Component	File	Folder
Text encoder	qwen3vl_4b_int8_convrot.safetensors	models/text_encoders/
FLUX.2 VAE	fibo2-flux2-vae.safetensors	models/vae/

Qwen3-VL 4B text encoderqwen3vl_4b_int8_convrot.safetensorsFLUX.2 VAE. Save as fibo2-flux2-vae.safetensorsflux2-vae.safetensors

The Edit Turbo workflow loads fibo2-test-input.png from ComfyUI/input/. Use the included sample or select another image in the workflow.

Sample edit inputfibo2-test-input.png

Workflow downloads

Import these JSON files into ComfyUI.

JSON fileFibo 2.0 Dev text-to-image workflowJSON fileFibo 2.0 Edit Turbo image-to-image workflow

License

The Dev and Edit Turbo model cards use CC BY-NC 4.0. Commercial use requires a separate agreement with Bria. Read the Dev model card and Edit Turbo model card for the full terms.