Independent guide. Cicadas does not run FLUX 3 Image Klein or represent Black Forest Labs.
FLUX Klein: FLUX 3 Image guide
FLUX 3 Image Klein is Black Forest Labs’ open-weights image model for generating pictures, editing with reference images and placing elements with bounding boxes. This guide points to its current downloads and setup instructions. It is separate from the FLUX video tool on Cicadas.
One image model
The official repository describes a 5.6B model for text-to-image, reference-based editing and layout control. It is an image checkpoint, not a video-generation preset.
Base or distilled
The distilled checkpoint uses fewer sampling steps; the Base checkpoint is intended for training and research. Choose by workflow rather than treating a developer speed claim as a measured result on your GPU.
Keep the prompt upsampler
The developer says the model expects captions from its prompt upsampler. Keep that step in the documented pipeline instead of sending a short raw prompt and assuming the same behavior.
How to start with FLUX Klein
- Choose the FLUX 3 Image Klein checkpoint on the official model page and read the weights license.
- Follow the reference repository or its ComfyUI instructions for the checkpoint, precision and memory available on your hardware.
- Use the prompt upsampler as documented. Save the model version and settings, then inspect your generated or edited picture.
Which FLUX Klein does this page cover?
This page covers FLUX 3 Image Klein. Search results also contain the earlier FLUX.2 [klein] family; use the exact checkpoint name when downloading or following a workflow. The existing FLUX 3 video page on Cicadas is a different media operation.
Read the current FLUX 3 Image Klein page ↗Editing and layout control
The reference repository documents editing with up to ten reference images and optional bounding boxes. Follow its reference ordering and box format. A workflow that loads the weights but drops these inputs does not demonstrate those controls.
Read the editing and layout examples ↗The prompt upsampler is part of the workflow
The official CLI and demo enable the upsampler by default. The README warns against running the model without it unless the input is already an upsampler caption. If you compare checkpoints, keep this part of the workflow consistent.
Read the prompt-upsample instructions ↗Check the weights license, not only the code license
The reference-code repository is labeled Apache-2.0, while the model weights link to the FLUX Kommunity License. They are different documents. Read the license attached to the exact checkpoint before planning commercial use or offering a hosted service.
Read the model weights license ↗FLUX Klein questions
Can I use FLUX Klein to generate an image here?
No. This is an independent model guide. The official model and setup links lead to the developer’s destinations; no FLUX Klein image job or credit charge is offered on this page.
Is FLUX Klein the same as FLUX 3 video?
No. This guide covers FLUX 3 Image Klein for still images and editing. The FLUX video page offers a separate video operation with its own model, input requirements and quote.
Does open weights mean free unlimited hosting?
No. Download access, license terms, local compute and hosted API fees are separate questions. Check the exact checkpoint license and the destination’s current terms; this page does not quote an image API price.
Should I choose Base or distilled?
Start from the workflow documented for your task. The distilled release is geared toward generation and editing; Base is the developer’s option for training, adaptation and research. We have not benchmarked either on Cicadas.
Can I skip prompt upsampling?
The official README says to use the prompt upsampler. Follow its exception only when your input is already an upsampler caption; a raw short prompt is not the same input.
Related image and video pages
By the Cicadas team · sources checked 2026-10-21 · · Testing and sources