Flux2-Klein-9B-Consistency¶

Development line: project:flux2-klein-9b-consistency · thread flux2-klein-9b-consistency-releases
Last event: 2026-03-13 · 1 dated since 2026-03-13 · Researched: 2026-09-05 · confidence: medium

What it is¶

Flux2-Klein-9B-Consistency is an image-to-image LoRA for base model black-forest-labs/FLUX.2-klein-9B.

• Source composition and detail preservation during semantic editing.
• Reduced color shift and oversaturation in V2.

The adapter is not an independent 9B model; base model license terms apply separately. Use V2 for controlled source image editing, not as a general generator.

Development line¶

• 2026-03-13 — Flux2-Klein-9B-Consistency resource linked. On 2026-03-13, the Flux2-Klein-9B-Consistency line linked to a Hugging Face resource under dx8152 and to a YouTube video. The links mark a public availability milestone without establishing model capabilities, version details, or intended workflow.

What changed¶

• 2026-03-13 — Released the Consistency adapter for FLUX.2 Klein 9B; a third-party entry from 2026-03-06 specified the workflow graph and CFG=1.
• 2026-04-17 — Added Flux2-Klein-9B-consistency-V2.safetensors to fix color tint, excessive detail, and oversaturation in V1. The V2 file is 331 MB with SHA-256 61db2017ce420b97bd5ef11984e5a894c90003a6bbf0dc9473f8d7b9ebb3ff93.

How to use this¶

Check the dx8152 Hugging Face repository and linked video from 2026-03-13 to evaluate fit before adoption.

1. Install current diffusers, transformers, and accelerate; load base black-forest-labs/FLUX.2-klein-9B in bfloat16 on CUDA. — https://huggingface.co/dx8152/Flux2-Klein-9B-Consistency
2. Load adapter weights via load_lora_weights("dx8152/Flux2-Klein-9B-Consistency") and pass the source image with edit instructions. — https://huggingface.co/dx8152/Flux2-Klein-9B-Consistency
3. For local ComfyUI, place the base 9B model, Qwen 3 8B text encoder, and flux2 VAE in their respective directories, then apply the LoRA in an image-edit workflow. — https://docs.comfy.org/tutorials/flux/flux-2-klein

Best practices¶

• Prefer V2 over V1: it addresses color shift, noisy detail, and oversaturation. — https://huggingface.co/dx8152/Flux2-Klein-9B-Consistency
• Start at strength 0.3–0.5 when combining with lighting LoRAs: this protects details but dampens the lighting shift. — https://www.patreon.com/foxfuressence/posts/lighting-angle-162733880
• Test against your base model version and workflow: user reports diverge, with poor results reported on FP8 and GGUF variants. — https://www.reddit.com/r/comfyui/comments/1rv6juw/fixing_the_plastic_look_in_flux2_klein_9b_with/

Superseded by this¶

• 2026-04-17 — V1 is obsolete whenever neutral color, clean detail, and controlled saturation matter.

Still unknown¶

• Primary logs do not confirm the exact upload date for V1; records from 2026-03-06 confirm the base model, purpose, and CFG=1 for the 2026-03-13 release.
• Video JXMbbbdfnSg was unreadable, leaving its contents unused as evidence.
• No independent reproducible benchmark confirms the claimed consistency gains; user reports diverge on quality and compatibility.
• The adapter repository lists Apache-2.0, whereas BFL documentation specifies the FLUX Non-Commercial License for 9B Base; verify base model weights before commercial use.

Sources¶

Agent brief¶

• Subject: project:flux2-klein-9b-consistency, thread flux2-klein-9b-consistency-releases, 1 dated events 2026-03-13 → 2026-03-13.
• Practical note: As of 2026-03-13, practitioners should recognize Flux2-Klein-9B-Consistency as a publicly linked dx8152 resource and inspect its Hugging Face page and linked video before deciding whether it fits their workflow; its capabilities and compatibility remain unverified.
• Confidence: medium. Dated supersedes above are the authority for what is obsolete.