<h3 id="about-these-variants">About these variants</h3>Four precision levels are available for FLUX 3 Image Klein (available for both the Distilled and Base checkpoints), trading VRAM and speed against generation quality: - BF16 — full-precision reference (~11.1 GB). Highest quality, largest footprint. Available on the official black-forest-labs/flux-3-image-klein-comfy repository. - INT8 (W8A8) ConvRot — 8-bit weights + 8-bit activations (~5.8 GB). Near-lossless quality, ~2× smaller than BF16, runs on INT8 tensor cores (RTX 30-series and up). - INT6 (W6A8) — 6-bit weights + 8-bit activations (~4.4 GB). Middle-ground: smaller than INT8 with only a small quality trade-off. - INT4 (W4A8) — 4-bit weights + 8-bit activations (~3.2 GB). Smallest footprint; uses ConvRot with a per-tensor codebook that decodes to INT8 for compute, so it runs on the same INT8 hardware as W8A8. Larger quality trade-off than INT6. Two checkpoint flavors available at each precision, same architecture, different intended use: - Klein Distilled (flux3_image_klein_sd_*) — step-distilled for fast low-step inference. The default for day-to-day image generation and editing. Served as "FLUX 3 Image Klein" on the BFL API. - Klein Base (flux3_image_klein_*) — non-distilled base checkpoint for fine-tuning, LoRA training, and research. Use this as the starting point for training adapters or further post-training; not optimized for few-step inference out of the box. FLUX 3 Image Klein — INT8 / INT6 / INT4 ConvRot for ComfyUI Quantized weights of FLUX 3 Image Klein for fast, low-VRAM inference in ComfyUI. These are modified (quantized) versions of the official FLUX 3 Image Klein checkpoints from Black Forest Labs, repackaged for ComfyUI with mixed-precision quant metadata that loads via the stock Load Diffusion Model node. The BF16 reference is linked for comparison but is available directly from Black Forest Labs' official repository. About FLUX 3 Image Klein A unified text-to-image generation and image editing model in the FLUX family, released October 15, 2026. With 5.6B parameters in the image transformer (text encoder separate), Klein balances generation quality, inference efficiency, and versatility, and shares the FLUX 3 foundation underlying BFL's larger FLUX 3 Image and FLUX 3 Action releases. Highlights- Efficient Image Generation — strong visual quality at the 5.6B class with a step-distilled variant for low-step inference, making high-quality image creation accessible across a wide range of creative workflows. - Flexible Creative Control — unified generation and editing in a single checkpoint, with bounding-box conditioning to specify where elements appear in the frame. Key features - Unified Generation and Editing — generate from text and edit source images in the same model, no separate edit checkpoint required. - Bounding-Box Conditioning — specify where elements should appear in the output by providing bounding boxes alongside the prompt. - Qwen3-VL-4B Text Encoder — multimodal-capable text encoder for stronger prompt grounding than CLIP-class encoders. - Distilled and Base Variants — one distilled checkpoint for fast inference, one base checkpoint for fine-tuning and LoRA training.<h3 id="license">License</h3>FLUX 3 Image Klein is licensed by Black Forest Labs Inc. under the [FLUX Kommunity License](<a target="_blank" rel="ugc" href="https://bfl.ai/legal/flux-kommunity-license">https://bfl.ai/legal/flux-kommunity-license</a>). Copyright Black Forest Labs Inc. IN NO EVENT SHALL BLACK FOREST LABS INC. BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH USE OF THIS MODEL. Free commercial use is permitted for individuals and businesses with annual revenue under $5M USD; larger organizations should contact Black Forest Labs for commercial licensing.

Files with SHA256: 8949ecfc1bf3871b08be02cf43ec7276a4914e1a86eade6b5eb8f4204426bc92