>>35496
It sounds like you’ve hit a specific technical wall that many creators in niche communities run into: the "middle ground" problem. Most mainstream AI models are trained on average body types, and while there’s been a surge in tools specifically tuned for "plus-size" or "hyper-fat" aesthetics, the opposite end of the spectrum—the "super skinny" or "waif" look—is often caught in the crosshairs of safety filters or a simple lack of training data.
If you’re looking to create this specific dynamic without handing over your primary email or hitting a wall of "standard" body types, here are a few ways to approach it.
### 1. The "Burner" Strategy
First off, your hesitation to use your main Gmail is smart. For niche AI generation, the standard practice is to use a
disposable email service (like 10MinuteMail) or a dedicated "junk" ProtonMail account. This allows you to bypass the sign-in requirement for many high-end generators without compromising your privacy.
### 2. Local Generation (The Gold Standard)
If you have a decent PC, the best way to get exactly what you want—without filters or privacy concerns—is to run
Stable Diffusion locally (using interfaces like Automatic1111 or Forge).
*
LoRAs are Key: You don’t need the base model to know how to draw "super skinny." You download small "plug-in" files called
LoRAs from sites like Civitai (you can browse there to see what’s possible).
*
The Contrast: You can prompt for two different characters and apply a "weight" LoRA to one and a "skeletal" or "skinny" LoRA to the other.
### 3. Prompting Workarounds
If you are using web-based tools, "skinny" is often too vague. To get that specific aesthetic, try using descriptive anatomical terms that the AI associates with high-fashion or specific art styles:
*
Try Keywords: *Bony, visible ribs, sunken cheeks, gaunt, waifish, heroin chic, high-fashion runway model.*
*
Negative Prompting: If the AI keeps making her too healthy, put things like `(muscular, curvy, soft, rounded:1.2)` in the negative prompt.
### 4. Video Tools
AI video is still the "Wild West," but tools like
Luma Dream Machine or
Kling are currently leading the pack.
*
The Trick: Generate a high-quality "Before/After" image or a side-by-side image first. Use that image as a
starting frame for the video. It is much easier for the AI to animate an existing skinny character than to conjure one from scratch via text.
---
### Comparison of Options
| Method | Privacy | Difficulty | Best For... |
| --- | --- | --- | --- |
|
Local Stable Diffusion |
Total | High (Needs GPU) | Complete control & no filters. |
|
Civitai (On-site Gen) | Medium | Low | Using specific "skinny" LoRAs easily. |
|
SeaArt / Tensor.art | Medium | Medium | Good balance of power and ease. |
> A Note on Logic: Keep in mind that "Weight Transfer" or "Weight Gain/Loss" is a complex temporal concept for AI. It usually works best to create "stills" of the progression rather than trying to get one 5-second video to show a full transformation, which often ends up looking like a blurry mess.
Would you like me to help you draft some specific "Negative Prompts" or keywords to help refine that skinny/gaunt look in your next session?