PULPKHOR.safetensors

79 MB · SafeTensor

2712368_training_data.zip

59 MB · Other

Memerlukan keahlian

<retro-comic-flux2>

Over the last two weeks I’ve been experimenting a lot with Qwen Image Edit 2511 and FLUX.2 Klein, trying to better understand how these models behave when training style LoRAs. That’s the main reason I haven’t been uploading many styles recently. Most of my time went into testing, failing, tweaking parameters, and learning how these systems actually respond. This LoRA is still a beta. My goal with it was to push a pulp / vintage comic look with real texture and artifacts, without just overlaying a filter. It took many attempts and different setups to get something that actually affects structure, shading, and surface detail instead of just colors. (note: it works better for editing human portraits, for 2d illustrations it adds some texture and can change colors) I’m still learning how to properly tune things like learning rate, steps, repeats, schedulers, and rank for this kind of training, especially on these newer models. So I’m very open to feedback and tips from people who have more experience with this. I’m also sharing my dataset so you can check the captions and structure yourself. If you have suggestions on how to improve the training process for this kind of style, I’d really appreciate it. Thanks to everyone who takes the time to test it and give feedback.

SHA256

52F4A77C051AD349A71BC6174983F0BDC25E10A19FECC82E6A874B578953705E

ID Model ModelTrends.ai

#24827

Saiz

672x1024

Saiz

1024x1280

Sampler

Euler

Langkah

4

Guidance

1

Saiz

2048x1024

Sampler

Euler

Langkah

12

Guidance

1

Seed

1060638152

Saiz

1216x832

Berat yang digunakan

1

Sampler

Euler

Langkah

12

Guidance

1

Seed

2216032129

Saiz

1216x832

Berat yang digunakan

1

Saiz

672x1024