Hi, I’m using generateImage (text-to-image-plugin) for an 18+ character generator, and I keep hitting the model’s limits. Three concrete requests:

1. Enable reference images (img2img). The plugin’s referenceImage option is in the code but disabled server-side. Text alone can’t match a specific body type / render style — being able to pass a reference image (even with a strength/blur slider) would be the single biggest improvement.

2. A higher-capability model / higher resolution. The current one caps at 512–768px and drops details on compound prompts. 1024+ and better prompt adherence would help enormously.

3. Better prompt fidelity. The model over-weights the first ~100 words, so long prompts get diluted. Anything like emphasis syntax ((word:1.2)), better token weighting, or a higher max guidance would help.

Context: I’m generating exaggerated-curvy “Daz render style” characters. Explicit content sometimes gets refused or softened by the model’s filter — I understand that’s by design, just noting it.

I want my generated characters to have bodies that match images on this DeviantArt page: https://www.deviantart.com/balloondolls