Fibo 2.0
Fibo 2.0 (Dev) needs 9.4 GB of VRAM in ComfyUI, at the prompt enhancer of its own template, by file sizes read 3 November 2026. The smallest card that loads it with room to work is the RTX 5070 (12 GB), and the template downloads 27.0 GB.
- Made by
- Bria AI
- Licence
- CC BY-NC 4.0
- Released
- October 2026
- Sizes read
- 3 November 2026
Largest stage from exact file sizes, with ComfyUI's own template; LoRAs are not counted. Working memory for the generation comes on top. Open it in the ComfyUI checker
Largest stage from exact file sizes, with ComfyUI's own template; LoRAs are not counted. Working memory for the generation comes on top. Open it in the ComfyUI checker
Which file on your card
For each memory size, the best file for that card: the highest-precision diffusion file that loads with room to work (one that only just fits when nothing else does), and what is left for working memory while it samples. On a larger card that can be above the file ComfyUI's template loads, which the answer above is about. "Streams" means even the smallest file is larger than the card: ComfyUI still runs it from system RAM, more slowly. FP4 files go only to the Blackwell cards that run them natively.
Dev
ComfyUI's template for Dev: INT8 ConvRot diffusion model, BF16 text encoder. Stages: text encoder 8.9 GB, prompt enhancer 9.4 GB, diffusion 8.8 GB. Download 27.0 GB.
Computed 3 November 2026Edit Turbo
ComfyUI's template for Edit Turbo: INT8 ConvRot diffusion model, BF16 text encoder. Stages: text encoder 8.9 GB, prompt enhancer 9.4 GB, diffusion 8.8 GB. Download 27.0 GB.
Computed 3 November 2026Check another card, or pick a file by hand, in the ComfyUI VRAM checker. Built your own workflow around it, with LoRAs or custom nodes? Drop it into the ComfyUI workflow checker.
Where to run it
Dev
On-demand prices before tax, read 19 October 2026.
Edit Turbo
On-demand prices before tax, read 19 October 2026.
About the model
Fibo 2.0 is Bria AI's open image model. The second generation of Bria AI's FIBO, from October 2026: an 8-billion-parameter model that generates and edits images with one set of weights, from a prompt and up to four reference images. Dev is the full model, run with guidance at 50 steps; Edit Turbo is distilled to four steps and only edits.
Dev: 8.1B · Edit Turbo: 8.1B Text encoder: Qwen3-VL 4B CC BY-NC 4.0 Released October 2026
Before you download
Non-commercial use only, with credit to Bria AI; commercial use needs a separate agreement with Bria AI. Read the CC BY-NC 4.0 itself before using it for work.
Good at
- Instruction edits, masked edits and multi-reference compositions from one checkpoint.
- Fast edits: Edit Turbo needs four steps and no classifier-free guidance.
- INT8 files from Comfy-Org at about half the size of BF16.
- A single-GPU LoRA trainer in Bria's own repo.
Watch out for
- CC BY-NC 4.0: the weights are for non-commercial use, with credit to Bria AI, and any commercial use needs a separate agreement with Bria.
- Stay near 1 megapixel, the size it was trained at: Comfy's template notes say prompt adherence collapses above it.
- ComfyUI's two local templates are for editing and run a Qwen3-VL 8B prompt rewriter by default. ComfyUI has no guide to running it locally.
- Bria's own repos are gated; Comfy-Org's repackaged files are not. The templates need ComfyUI 0.43.0 or later.
Every Fibo 2.0 file ComfyUI loads
Sizes are the exact byte counts Hugging Face reports, read 3 November 2026. Files marked "template" are the ones ComfyUI's official workflow downloads.
| Part | Precision | Size | Version | File |
|---|---|---|---|---|
| Diffusion model | BF16 | 16.45 GB | Dev | fibo2_dev_bf16.safetensors |
| Diffusion model | INT8 ConvRot · template | 8.41 GB | Dev | fibo2_dev_int8_convrot.safetensors |
| Diffusion model | BF16 | 16.45 GB | Edit Turbo | fibo2_edit_turbo_bf16.safetensors |
| Diffusion model | INT8 ConvRot · template | 8.41 GB | Edit Turbo | fibo2_edit_turbo_int8_convrot.safetensors |
| Text encoder | BF16 · template | 8.88 GB | all | qwen3vl_4b_bf16.safetensors |
| Text encoder | INT8 ConvRot | 4.86 GB | all | qwen3vl_4b_int8_convrot.safetensors |
| Prompt enhancer | INT8 ConvRot · template | 9.35 GB | all | qwen3vl_8b_int8_convrot.safetensors |
| VAE | FP32 · template | 0.34 GB | all | flux2-vae.safetensors |
What Bria AI says it needs
Bria states no VRAM figure on either model card, and ComfyUI has no guide to running it locally, only its two templates. The closest it comes: "Fibo 2.0 is an 8B-parameter flow-matching model operating in the latent space of the FLUX.2 VAE." (source).
Those figures and this page measure different things. A vendor's figure covers its own script at its default resolution, working memory included. This page counts the weights each stage holds, which is exact, and leaves the working memory as the space that remains, because it depends on your resolution and attention kernel and nobody publishes it per model. If generation runs out of memory with a file that loads, lower the resolution before stepping down a precision. ComfyUI's README says it "can run even the biggest open source models on as low as 4GB vram + 8GB ram" by streaming weights (source), so "streams" means slower, not impossible.
LoRA training for Fibo 2.0
No trainer we checked publishes a VRAM figure for training Fibo 2.0 yet (kohya's sd-scripts and musubi-tuner, ostris' ai-toolkit, diffusion-pipe, and Bria AI's own repo). When one does, it goes here.
Questions
- How much VRAM does Fibo 2.0 need?
- In ComfyUI's own workflow for Dev, the largest stage is the prompt enhancer at 9.4 GB, and the whole workflow downloads 27.0 GB. Working memory for generation comes on top and grows with resolution. Bria AI publishes no VRAM figure.
- Can I run Fibo 2.0 on 8 GB of VRAM?
- Not fully. The smallest file here, INT8 ConvRot at 8.4 GB, needs 9.4 GB at its largest stage, more than 8 GB holds. ComfyUI will still run it by streaming weights from system RAM, which works but is slower.
- Can I run Fibo 2.0 on 12 GB of VRAM?
- Yes, with the INT8 ConvRot file: its largest stage, the prompt enhancer, is 9.4 GB, which loads on 12 GB with 3.2 GB left for working memory.
- Can I run Fibo 2.0 on 16 GB of VRAM?
- Yes, with the INT8 ConvRot file: its largest stage, the prompt enhancer, is 9.4 GB, which loads on 16 GB with 7.2 GB left for working memory.
- Which Fibo 2.0 file should I download?
- The highest precision that loads fully on your card. On a 24 GB card that is the BF16 file (16.4 GB). ComfyUI's template downloads INT8 ConvRot with the BF16 text encoder. FP4 files (NVFP4) are made for Blackwell cards (RTX 50, RTX PRO 6000, DGX Spark); on older cards, use FP8 or INT8.
- Can I use Fibo 2.0 commercially?
- Not under its open licence. The weights are released under CC BY-NC 4.0, for non-commercial use with credit to Bria AI, and Bria sells commercial use through a separate agreement.
- Files and sizes: Comfy-Org/Fibo-2.0 on Hugging Face, read from the Hugging Face API. Template files: ComfyUI's Fibo 2.0 template.
- Model and licence: briaai/Fibo-2.0-dev and the CC BY-NC 4.0.
- Card memory: manufacturers' specifications, as on each GPU page. The checker covers NVIDIA cards; ComfyUI also runs on AMD and Apple silicon, where file formats and speed differ and we have not sized them.
- Fit rule: a stage loads fully under 85% of usable memory and is tight up to 95%, the same margins the LLM pages use.
- Prices: each provider's own pricing page, on the dates beside them.