DogePoint Collection - PDXL版本DogePONY - Draw V1 (ID: 991033)

Update v1.25: Contains some small fixes and features added, preferably over the 1.2. I won't further work on it until I can list it on site generation or I get some relevant feedback.

DOGEPONY 2D:

This model should be pretty close to the original Pony V6 and you can probably recreate most of the stuff with the same seeds very closely. The main differences are: more realistic and 3D data, more anime data, less deviation/sharper, some LoRAs with unusual resolutions and the tweaked TE that might be a problem if you repeat yourself too much in the prompt.

For sampling, there's no skeleton key, try to keep awareness of the content you're trying to generate. DPM samplers paired with Karras schedulers can give you some rough, painterly shading at low steps and more realism, details at high steps, white Euler and DDPM can give you some nice flat styles and more 3D at higher steps. You can always check the showcased LoRAs prompt for details.

HiresFix and Adetailer highly recommended and used in the example pics. If you run into problems using multiple LoRAs, try toning down a bit on the weight or CFG rescale.

描述:

Tested this version with multiple resolutions, different scripts and it seemed to go swimmingly through the most of it. Would appreciate your feedback with posted images or reports on what might be broken.

What's new:

• Better tuning

• Less concept/color bleeding

• Less disruptive with LoRAs and styles

• Adjusted NSFW

• Less realistic and 3D interference

• Sharper and cleaner faces

• More noise for better and detailed scenery/ backgrounds

Known Issue:

• some baked characters might be too strong and disruptive

If you run into any trouble, try using Dynamic Thresholding or LoRA for said characters.

训练词语: