Generate an image · Your LoRA file
Qwen Image 2.1 LoRA · Text to Image
- loras
- Up to 3 files · Strength 0–4
Start with the author’s recommended strength.
Open this workflowAI image & video LoRA
Find the right model for your LoRA.
A great character or style LoRA needs the right base. Match its training model, exact version and your image or video task, then go straight to a compatible workflow.
Match the base. Pick the task.
Choose the Base Model listed on the LoRA’s model card. Results use registered training bases and each current endpoint’s LoRA schema; your file stays with you.
Use the Base Model on the LoRA’s model card. No upload is needed.
Matching workflows
Available workflows: 1
Which models take your LoRA?
Start here even before choosing a file. Image generation, editing and video can use different weights within the same model family. Links below open the actual task; only currently available, schema-confirmed entries appear.
A successful job is not the whole compatibility test.
These saved Qwen Image 2.1 examples make the difference easy to see: the Grainscape example has a distinct texture, while the Qwen 2512 example looks close to the image without a LoRA. Match the base first, then compare one strength at a time.
Saved outputs from the Qwen Image 2.1 LoRA guide.
From a downloaded LoRA to a useful first take
- 01
Read the training base, not just the name
Open the LoRA’s model card and find Base Model. Keep the exact edition: Qwen 2.1 is different from 2512, and Z-Image Base is different from Turbo.
- 02
Choose the job you are making
A character portrait, an edit and an animated opening frame are different tasks. Match the task here, then use the model’s actual LoRA slot and the author’s trigger phrase.
- 03
Compare a short, controlled draft
Save a no-LoRA baseline. Keep the prompt, source image and seed steady, add one matching LoRA, and vary its strength. For a NSFW character series, check face, outfit and framing before producing the full set.
LoRA compatibility, without the guesswork
Choose weights trained for the same architecture and edition. A shared file extension does not make Qwen, FLUX and Wan weights interchangeable. The checker gives you matching tasks and a direct route to change the target model.
It matches the training base and task you select against the current catalogue and input schemas. It does not download your file. Use the Base Model field on its model card, then open the matching Playground to try your weights.
Choose a Qwen Image 2.1 LoRA for the 2.1 workflow. For 2512 image generation, select the 2512 base; for the older editing workflow select Edit 2511. Jib Mix also lists compatible Qwen-Image/2512 weights for text-to-image.
Choose the Wan 2.2 training mode first: T2V for text-to-video, I2V for an opening image or continuation. Put paired files in high_noise_loras and low_noise_loras as directed by their author; the shared loras slot is for weights applied to both passes.
LTX 2.3 Spicy LoRA uses named presets from the current schema. Select a preset in that Playground. For your own weight files, choose the LTX-2.3 LoRA workflow for LTX-2.3 files or the LTX-2 19B LoRA workflow for 19B files.
Compatibility gives the weights the correct place to run. The visible result also depends on the training, trigger phrase, strength and your source image. Start with one consenting adult character and a controlled draft, then keep the take that fits your project.
Keep the style. Start with the right base.
Explore model-specific LoRA guides, saved examples and current workflows for character, style and motion.