FLUX 3 Image Klein
FLUX 3 Image Klein (Klein (distilled)) needs 9.8 GB of VRAM in ComfyUI, at the prompt enhancer of its own template, by file sizes read 16 October 2026. The smallest card that loads it with room to work is the RTX 5070 (12 GB), and the template downloads 20.8 GB.
- Made by
- Black Forest Labs
- Licence
- FLUX Kommunity License
- Released
- October 2026
- Sizes read
- 16 October 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.
Klein (distilled)
ComfyUI's template for Klein (distilled): INT8 ConvRot diffusion model, INT8 ConvRot text encoder. Stages: text encoder 4.9 GB, prompt enhancer 9.8 GB, diffusion 6.2 GB. Download 20.8 GB.
Computed 16 October 2026Klein Base
ComfyUI's template for Klein Base: INT8 ConvRot diffusion model, INT8 ConvRot text encoder. Stages: text encoder 4.9 GB, prompt enhancer 9.8 GB, diffusion 6.1 GB. Download 20.8 GB.
Computed 16 October 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
Klein (distilled)
On-demand prices before tax, read 16 to 19 October 2026.
Klein Base
On-demand prices before tax, read 19 October 2026.
About the model
FLUX 3 Image Klein is Black Forest Labs' open image model. Black Forest Labs' open FLUX 3 model from October 2026: a 5.6-billion-parameter image model in two versions, the step-distilled Klein that samples in 16 steps and serves Black Forest Labs' own API, and Klein Base, the undistilled one for fine-tuning and LoRA training. One checkpoint makes images from text, edits them, and composes from up to 10 reference images.
Klein (distilled): 5.6B · Klein Base: 5.6B Text encoder: Qwen3-VL 4B FLUX Kommunity License Released October 2026
Before you download
The weights are for non-commercial use; people and companies under US$5 million in yearly revenue, affiliates included, may use the images they generate commercially, with content filtering or review, and labelled as AI-made where the law requires. Anything else needs a licence from Black Forest Labs. Read the FLUX Kommunity License itself before using it for work.
Good at
- Text to image, edits and multi-reference compositions without loading a second model.
- Layout by bounding boxes in the prompt: place, move, recolour or remove elements on a 0 to 1000 grid.
- int8 and NVFP4 files from Black Forest Labs at about half the size of BF16; the int8 ones also run on cards without FP8 support.
- Commercial use of the images for people and small companies under US$5 million in yearly revenue.
Watch out for
- The FLUX Kommunity License is not an open-source licence. The weights are for non-commercial use, commercial use of the images is only for those under US$5 million in yearly revenue, and that use needs content filtering or review, plus an AI label where the law requires one. Anything else needs a licence from Black Forest Labs.
- Black Forest Labs recommends its prompt upsampler for every prompt, and ComfyUI's text-to-image templates load it: a separate 9-billion-parameter language model, larger than the image model. The editing templates have no upsampler, so write the full caption there yourself.
- The files sit in a gated Hugging Face repo. Accept the licence there, signed in, before ComfyUI's download links work.
- Black Forest Labs' figures measure different things: the model card's peaks are for its reference code on a B200, while the ComfyUI repo's card says the int8 files run on 8 GB cards.
- It needs ComfyUI 0.41 or later, and the NVFP4 files need a Blackwell GPU.
Every FLUX 3 Image Klein file ComfyUI loads
Sizes are the exact byte counts Hugging Face reports, read 16 October 2026. Files marked "template" are the ones ComfyUI's official workflow downloads.
| Part | Precision | Size | Version | File |
|---|---|---|---|---|
| Diffusion model | BF16 | 11.10 GB | Klein (distilled) | flux3_image_klein_sd_bf16.safetensors |
| Diffusion model | INT8 ConvRot · template | 5.83 GB | Klein (distilled) | flux3_image_klein_sd_int8_convrot.safetensors |
| Diffusion model | NVFP4 | 5.28 GB | Klein (distilled) | flux3_image_klein_sd_nvfp4.safetensors |
| Diffusion model | BF16 | 11.08 GB | Klein Base | flux3_image_klein_bf16.safetensors |
| Diffusion model | INT8 ConvRot · template | 5.81 GB | Klein Base | flux3_image_klein_int8_convrot.safetensors |
| Diffusion model | NVFP4 | 5.26 GB | Klein Base | flux3_image_klein_nvfp4.safetensors |
| Text encoder | BF16 | 8.88 GB | all | qwen3vl_4b_bf16.safetensors |
| Text encoder | INT8 ConvRot · template | 4.86 GB | all | qwen3vl_4b_int8_convrot.safetensors |
| Prompt enhancer | INT8 ConvRot · template | 9.75 GB | all | qwen3.5_9b_flux3_prompt_upsampler_int8_convrot.safetensors |
| VAE | FP32 · template | 0.34 GB | all | flux2-vae.safetensors |
What Black Forest Labs says it needs
- "Peak GPU memory with the reference code, B200, 1024x1024, prompt upsampler off: 31 GB in BF16, 26 GB in FP8, 25 GB in NVFP4." (Black Forest Labs' reference code on a B200 at 1024×1024 in BF16, prompt upsampler off, source)
- "The prompt upsampler adds about 28 GB on an 80 GB GPU." (the prompt upsampler in the reference code, on top of the model, source)
- "The int8 files run on 8 GB GPUs such as the RTX 3070, and on GPUs without FP8 support (RTX 30 series and older)." (the int8 files, the card of Black Forest Labs' ComfyUI repo, 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 FLUX 3 Image Klein
No trainer we checked publishes a VRAM figure for training FLUX 3 Image Klein yet (kohya's sd-scripts and musubi-tuner, ostris' ai-toolkit, diffusion-pipe, and Black Forest Labs' own repo). When one does, it goes here.
Questions
- How much VRAM does FLUX 3 Image Klein need?
- In ComfyUI's own workflow for Klein (distilled), the largest stage is the prompt enhancer at 9.8 GB, and the whole workflow downloads 20.8 GB. Working memory for generation comes on top and grows with resolution. Black Forest Labs itself says: "Peak GPU memory with the reference code, B200, 1024x1024, prompt upsampler off: 31 GB in BF16, 26 GB in FP8, 25 GB in NVFP4." (Black Forest Labs' reference code on a B200 at 1024×1024 in BF16, prompt upsampler off).
- Can I run FLUX 3 Image Klein on 8 GB of VRAM?
- Not fully. The smallest file here, INT8 ConvRot at 5.8 GB, needs 9.8 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 FLUX 3 Image Klein on 12 GB of VRAM?
- Yes, with the INT8 ConvRot file: its largest stage, the prompt enhancer, is 9.8 GB, which loads on 12 GB with 5.8 GB left for working memory.
- Can I run FLUX 3 Image Klein on 16 GB of VRAM?
- Yes, with the BF16 file: its largest stage, the diffusion stage, is 11.4 GB, which loads on 16 GB with 4.6 GB left for working memory.
- Which FLUX 3 Image Klein file should I download?
- The highest precision that loads fully on your card. On a 24 GB card that is the BF16 file (11.1 GB). ComfyUI's template downloads INT8 ConvRot with the INT8 ConvRot 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 FLUX 3 Image Klein commercially?
- Only in part. The FLUX Kommunity License allows non-commercial use, and lets people and companies under US$5 million in gross annualised revenue, affiliates included, use the images they generate commercially, as long as they filter or review the output and label it as AI-made where the law requires. Commercial use of the model itself needs a licence from Black Forest Labs.
- Files and sizes: black-forest-labs/flux-3-image-klein-comfy on Hugging Face, read from the Hugging Face API. Template files: ComfyUI's FLUX 3 Image Klein guide.
- Model and licence: black-forest-labs/flux-3-image-klein and the FLUX Kommunity License.
- 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. Comfy Cloud's answer comes from ComfyUI's own template for that variant, which uses built-in nodes only, and shows only for the variants whose template we checked.