Uncensored Fibo 2.0 Edit Turbo: BRIA's Fast Open-Weights Image Editor

Where Edit Turbo sits in BRIA's Fibo 2.0 line

This Turbo checkpoint is the distilled editing model in BRIA AI's Fibo 2.0 family. Its Hugging Face card, briaai/Fibo-2.0-edit-turbo, names the Dev model as the base model, and both share the same 8B architecture: Qwen3-VL-4B-Instruct reads the instruction and images, a Perceiver resampler compresses them, a single-stream DiT does the work and the FLUX.2 VAE decodes the result. Distillation cuts sampling to 4 steps at guidance 1.0, which is what makes it suited to interactive tools and batch retouching. It cannot generate from text alone. BRIA offers Fibo 2 as a whole through open weights, its hosted platform, self-hosting and partners such as fal; the API is only one way to use it. Prompts should be written in English, the only language listed on the card.

Three ways to edit with Edit Turbo

The model accepts the same edit inputs as BRIA's v2 edit endpoint, so the limits documented there are a good guide for any front end built on it:

• Instruction edit: one image plus a sentence such as "make the jacket cherry red"; areas you do not mention are meant to stay as they were.
• Mask edit: one image plus a mask of the same size; black is kept, white is regenerated, and the output keeps the input resolution. Masks cannot be combined with several images.
• Multi-reference edit: 1 to 4 images on BRIA's API (more than 4 returns an error; the diffusers pipeline allows 5), for example a model photo, a garment and a location.

Where a fast editor pays off

Four-step editing changes what is practical. Retouching that used to mean a queue becomes something a user can drag and retry. Early adopters are mostly ComfyUI users: the official repository had 68 likes and 310 downloads in its first two days, and Comfy-Org repackaged it with bf16 and INT8 files. Typical jobs that suit it:

• Catalog fixes: recolor one product across hundreds of shots without reshooting.
• Background swaps for listings and social posts while the subject stays intact.
• Interactive editors and plugins where each click needs a quick preview.
• Ad variants: the same model photo in several outfits or settings for testing.

Edit Turbo vs Dev: speed or full control

Pick the model by the job. Turbo trades flexibility for speed: four steps instead of thirty or more, no text-to-image, and according to the model card summary a negative prompt has no effect. Dev is slower but generates and edits, honours negative prompts and is the one to fine-tune. A sensible workflow uses Turbo for quick previews and routine fixes, then reruns hard edits such as large structural changes on Dev.

Edit prompt guide with templates

Turbo responds best to short, specific instructions. Name the element, the change and what must not move. Do not rely on a negative prompt; write the constraint into the instruction instead. Large changes of aspect ratio on a single image can degrade quality, so crop first or use several images. For multi-image edits, refer to each image by its order. Templates:

• Recolor: "Change the [item] to [color] [material]; keep the face, pose, lighting and background identical."
• Background: "Replace the background with [scene]; keep the person, outfit and shadows unchanged."
• Mask fill: "In the masked area, add [object or garment] matching the existing light direction."
• Combine: "Dress the person in image 1 in the outfit from image 2 and place them in the setting of image 3."
• Relight: "Add warm late-afternoon light from the right; do not change the composition."

Run Edit Turbo in ComfyUI

ComfyUI core added Fibo 2.0 support on October 21, 2026, one day after diffusers merged its BriaFibo2Pipeline. Accept the gated terms on Hugging Face, or use Comfy-Org's repackaged files, then set up the graph as follows. BRIA gives no official VRAM minimum; the INT8 files are the practical choice on 24GB cards or less.

• Put fibo2_edit_turbo_bf16 or fibo2_edit_turbo_int8_convrot in models/diffusion_models.
• Put qwen3vl_4b_bf16 or its INT8 version in models/text_encoders, and flux2-vae in models/vae.
• Feed the source images into the TextEncodeFibo2 node together with the instruction.
• Sample at 4 steps, CFG 1, euler with the simple scheduler; more steps do not help a distilled model.

Is Edit Turbo open source?

It is open weights, not open source. Download requires a Hugging Face login and acceptance of BRIA's gated terms, and the bria-fibo license points to CC BY-NC 4.0, which allows non-commercial use only. Using it in a paid product needs a separate agreement with BRIA, either through its hosted platform or a self-hosting license that covers both generation and editing. The model card does not spell out whether edited outputs from the open weights may be used commercially, so confirm with BRIA first.

Does the Turbo checkpoint have filters?

BRIA's hosted edit endpoint can moderate prompts, input images and outputs, and the Fibo line is trained on licensed data that excludes nudity, violence and public figures. The downloadable checkpoint has no such moderation step, so locally it works as an uncensored Fibo 2.0 Edit Turbo. Photographers retouching boudoir, swimwear or glamour shots of consenting adults use it for that reason, though explicit detail is limited by the SFW training data and a distilled model is a poor base for retraining.

Yes, as open weights. Run in ComfyUI or diffusers, the Turbo checkpoint applies no prompt, input or output moderation; only BRIA's hosted endpoint does. It follows edit instructions on adult photos such as outfit or lighting changes, but its SFW training data makes explicit detail weak, and distilled models are hard to fine-tune without losing speed. As of October 21, 2026 no NSFW derivative had appeared on Hugging Face. The CC BY-NC license keeps any uncensored Fibo 2.0 Edit Turbo use non-commercial, and you should only edit photos of adults you have permission to change.

Uncensored photo editing on Spicy AI

For adult (18+) edits, Spicy AI's image editor is the direct option. It offers similar uncensored photo editing and works in the browser without downloads or a GPU. Upload a reference photo, paste one of the edit templates above, say what must stay unchanged and generate. Only edit photos you own or have permission to change, and never depict minors or real people without consent.

1. Keep instructions short: element, change and what must stay.
2. Write constraints into the instruction instead of a negative prompt.
3. Crop to the final ratio before editing a single image.
4. State that everyone shown is an adult (18+), and edit only photos you have the right to use.

Sources

• briaai/Fibo-2.0-edit-turbo model card (Hugging Face)
• Fibo 2 release notes (BRIA docs)
• Image edit API reference (BRIA docs)
• Safety and transparency (BRIA docs)
• Comfy-Org Fibo 2.0 repackaged files (Hugging Face)
• Fibo 2.0 support pull request (ComfyUI)
• Bria Fibo 2 pull request (diffusers)

Other uncensored image model guides

• Uncensored Midjourney
• Uncensored DALL·E
• Uncensored Stable Diffusion
• Uncensored Flux
• Uncensored Nano Banana
• Uncensored Nano Banana 2
• Uncensored Nano Banana 2.1
• Uncensored Nano Banana Pro
• Uncensored Ideogram
• Uncensored Leonardo AI
• Uncensored Seedream
• Uncensored Seedream 5.0
• Uncensored Krea
• Uncensored Kroma
• Uncensored Qwen-Image
• Uncensored Adobe Firefly
• Uncensored GPT Image 2.5
• Uncensored Fibo 2.0 Dev
• Uncensored GPT Image 2
• Uncensored Text to Image
• Uncensored Z-Image Turbo
• Uncensored Z-Image
• Uncensored FLUX.2
• Uncensored FLUX.2 [dev]
• Uncensored FLUX.2 [pro]
• Uncensored GPT Image 1.5
• Uncensored AI Head Swap
• Uncensored AI Cloth Swap