Consistence Edit — FLUX.2-klein-4B LoRA (mirror)
A rank-128 instruct-edit LoRA for FLUX.2-klein-4B that restores / adds realistic high-frequency detail to a degraded ("corrupted") input image while preserving consistency (color, structure) with the source. Consistency is controlled by LoRA strength — higher = more consistent, but less editing freedom.
This is a verbatim mirror, not original work. All credit for the weights belongs to the original author. This repository exists so downstream code can depend on a stable, long-lived URL; the file is byte-for-byte identical to the author's upload (verify via the SHA-256 below).
Attribution & source
- Original author: xiaozhijason (a.k.a. lrzjason on Hugging Face)
- Author's official release: https://huggingface.co/lrzjason/Consistance_Edit_Lora
(file
f2k_4B_consist_20260314.safetensors, published by the author under Apache-2.0) - Origin: CivitAI — Consistence Edit Lora, model
1939453, version2771678("Flux2 Klein 4B 20260314"): https://civitai.com/models/1939453?modelVersionId=2771678 - License: Apache-2.0, as released by the author for this Klein-4B version (declared
on both the CivitAI version page and the author's Hugging Face repo). Mirrored here under
those terms with attribution retained (see
NOTICE). If you are the author and want this mirror changed or removed, open a discussion and it will be actioned promptly.
Provenance (verify the mirror)
| File | f2k_4B_consist_20260314.safetensors |
| Size | 385,379,872 bytes (~367 MB) |
| SHA-256 | BED1E7E8747BC4572A20A7FCEF436E82CFBFA5430D0162EDA713B55D7A3C7CE4 |
| BLAKE3 | 66D285EA9535D1B97D445D847688681DE8F46AE54E8F63B61591B00B86CCD439 |
| Tensors | 200 (bf16) |
| Rank | 128 |
| Base model | black-forest-labs/FLUX.2-klein-4B |
| Key dialect | diffusers-split (transformer.…_transformer_blocks.N.…) |
Usage
- Base: FLUX.2-klein-4B (image edit).
- Suggested prompt:
{your edit instruction}. transform the image to realistic photograph. add realistic details to the corrupted image. restore high frequence details from the corrupted image. - Strength: 0.5–0.7 (per-image optimum varies).
- Applies to 60 double-stream + 40 single-stream projections (100 targets) under the Xocialize Klein LoRA applicator.
Disclaimer
Mirrored in good faith based on the author's Apache-2.0 release. No affiliation with or endorsement by the original author is implied.
- Downloads last month
- 30
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support
Model tree for xocialize/consistence-edit-FLUX.2-klein-4B-lora
Base model
black-forest-labs/FLUX.2-klein-4B