# FLUX.2 [klein] 4B Base

> FLUX.2 [klein] 4B Base is Black Forest Labs' 4-billion-parameter rectified flow transformer for text to image and multi-reference editing, trained without step or guidance distillation so it suits fine-tuning and LoRA work. Its weights are Apache 2.0 and need about 13 GB of VRAM. On Kenerate AI it costs 3 credits per image and 4.5 per edit with up to 3 references; the distilled FLUX.2 [klein] 4B (4 steps) is also 3 credits and is on the free plan.

URL: https://kenerateai.com/models/black-forest-labs/flux-2-klein-base-4b
Publisher: Kenerate AI (https://kenerateai.com)
Last updated: 20 October 2026

## Key facts
- Official name: FLUX.2 [klein] 4B Base
- Hugging Face ID: black-forest-labs/FLUX.2-klein-base-4B
- Maker: Black Forest Labs
- Size: 4B parameters · BF16
- Distillation: None (base model)
- Licence: Apache 2.0
- Local VRAM: ≈13 GB (RTX 3090 / 4070 and up)
- On Kenerate: 3 credits · edit 4.5
- References: Up to 3
- Max size on Kenerate: 1536×1536

## Steps
1. Open the model — Go to /app/image and pick Flux 2 Klein Base 4B (the button on this page selects it for you). Sign in free to get started.
2. Write the prompt — Describe medium, subject, setting, light and camera in plain sentences. Put exact on-image words in quotes.
3. Pick a size and seed — Choose one of 8 sizes up to 1536×1536 and fix the seed when you want to compare prompt changes.
4. Edit with references — Upload up to 3 images and say what changes and what stays — an edit costs 4.5 credits.

## In depth

### Which FLUX.2 [klein] should you pick?
Black Forest Labs ships each [klein] size twice: a distilled model tuned for speed and an undistilled Base for control and fine-tuning [1][2]. Here is how the three you can run on Kenerate compare.
- FLUX.2 [klein] 4B (distilled) — 4 steps, guidance 1.0, under a second on fast GPUs; Apache 2.0; 3 credits and on Kenerate's free plan.
- FLUX.2 [klein] 4B Base — 50 steps, guidance 4.0, more varied outputs; Apache 2.0; 3 credits, edit 4.5.
- FLUX.2 [klein] 9B Base — the larger undistilled model, about 29 GB of VRAM; non-commercial weights; 4.5 credits, edit 6.3.

### Running FLUX.2 [klein] 4B yourself vs in the browser
Locally, both 4B models need about 13 GB of VRAM in BF16 [1][2]. ComfyUI's workflows load the diffusion model with a Qwen 3 4B text encoder and the FLUX.2 VAE, and on an RTX 5090 it measured about 17 seconds per Base image against 1.2 seconds for the distilled 4B [3]. On Kenerate there is nothing to install: 3 credits per image, 8 sizes up to 1536×1536 and up to 3 references for edits.

### Prompting the 4B Base
Write full sentences: medium first (photo, watercolor, flat poster), then subject, setting, light and lens. Put any words that must appear in quotes and keep them short — Black Forest Labs notes that rendered text can be inaccurate [2]. Because the Base gives more varied results than the distilled model [1], fix the seed when you want to compare two prompts, and change it when you want new ideas.

References:
[1] black-forest-labs/FLUX.2-klein-base-4B — Hugging Face model card: https://huggingface.co/black-forest-labs/FLUX.2-klein-base-4B
[2] black-forest-labs/FLUX.2-klein-4B — Hugging Face model card: https://huggingface.co/black-forest-labs/FLUX.2-klein-4B
[3] FLUX.2 [klein] tutorial — ComfyUI docs: https://docs.comfy.org/tutorials/flux/flux-2-klein
[4] Flux (text-to-image model) — Wikipedia: https://en.wikipedia.org/wiki/Flux_(text-to-image_model)

## FAQ
Q: What is FLUX.2 [klein] 4B Base?
A: FLUX.2 [klein] 4B Base is a 4-billion-parameter rectified flow transformer from Black Forest Labs, published on Hugging Face as black-forest-labs/FLUX.2-klein-base-4B. It generates images from text and edits them with multiple reference images. “Base” means it was trained without step or guidance distillation, which keeps it flexible for fine-tuning, LoRA training and research.

Q: What is the difference between FLUX.2 klein 4B and FLUX.2 klein 4B Base?
A: Both are 4B models with Apache 2.0 weights and about 13 GB of VRAM. FLUX.2 [klein] 4B is distilled: the model card's example runs it in 4 steps at guidance 1.0, and Black Forest Labs says it can finish in under a second. The Base keeps full guidance (50 steps, guidance 4.0 in the example) and gives more varied outputs, so it is the better starting point for fine-tunes and LoRAs. On Kenerate both cost 3 credits per image.

Q: Is FLUX.2 klein 4B free?
A: The weights of both 4B models are free to download under Apache 2.0, but you need your own GPU. On Kenerate the distilled Flux 2 Klein (4B) is on the free plan, while Flux 2 Klein Base 4B is a paid model at 3 credits per image from a one-time credit pack, with no subscription.

Q: Can I use FLUX.2 klein 4B commercially?
A: Yes, per Black Forest Labs: the 4B [klein] weights are released under Apache 2.0 and the model card says outputs can be used for commercial purposes. The 9B [klein] models use the FLUX Non-Commercial License instead. Images you generate on Kenerate follow Kenerate's terms of service.

Q: How much VRAM does FLUX.2 klein 4B need?
A: About 13 GB, according to the model card, which names the RTX 3090 and RTX 4070 and up. ComfyUI measured 9.2 GB for the Base and 8.4 GB for the distilled model with its FP8 files on an RTX 5090. On Kenerate nothing runs on your machine.

Q: Can I use FLUX.2 klein 4B Base online without a GPU?
A: Yes. Kenerate AI runs it in the browser as Flux 2 Klein Base 4B: 3 credits per image, 4.5 per edit, up to 3 reference images and 8 sizes up to 1536×1536, with seed control. There is nothing to install.

Q: Can FLUX.2 klein 4B Base edit images?
A: Yes. The model card lists multi-reference editing. On Kenerate, upload up to 3 images to Flux 2 Klein Base 4B and describe the change and what must stay, for example “same product, swap the background for white marble”. Each edit costs 4.5 credits.

Q: FLUX.2 klein 4B Base vs 9B Base — which should I use?
A: The 9B Base has more than twice the parameters and needs about 29 GB of VRAM locally; its weights are non-commercial. The 4B Base needs about 13 GB and is Apache 2.0. On Kenerate the 9B costs 4.5 credits per image and the 4B costs 3. Pick the 4B Base when you plan to fine-tune for commercial use or want the cheaper model.

Q: What settings does the model card recommend?
A: The Diffusers example generates at 1024×1024 with 50 inference steps and a guidance scale of 4.0, and suggests CPU offload to save VRAM. For the distilled 4B the example uses 4 steps and guidance 1.0. On Kenerate the steps and guidance are set for you; you choose the prompt, size and seed.

## Related pages
- FLUX.2 [klein] 9B Base: https://kenerateai.com/models/black-forest-labs/flux-2-klein-base-9b
- Flux 2 (all versions): https://kenerateai.com/model/flux-2
- Z Image: https://kenerateai.com/model/z-image
- Open source AI image generator: https://kenerateai.com/open-source-ai-image-generator
- All models: https://kenerateai.com/model

## Sources
- black-forest-labs/FLUX.2-klein-base-4B — Hugging Face model card: https://huggingface.co/black-forest-labs/FLUX.2-klein-base-4B
- black-forest-labs/FLUX.2-klein-4B — Hugging Face model card: https://huggingface.co/black-forest-labs/FLUX.2-klein-4B
- FLUX.2 [klein] tutorial — ComfyUI docs: https://docs.comfy.org/tutorials/flux/flux-2-klein
- Flux (text-to-image model) — Wikipedia: https://en.wikipedia.org/wiki/Flux_(text-to-image_model)