HOME | DD

Description
It might not look it, but this was a fairly involved piece, with lots of editing and inpainting. This is one of those instances where I had a good idea of what I wanted to work towards and, as is usual with that, I eventually got frustrated when things weren't going as I envisioned, and gave up once the image was at an okayish state

As with a lot of ideas, I started off by blocking out a basic scene in Blender since I'm not great with visualising 3D spaces enough to just draw something freehand. Then, as a test to see how well it could work, I utilised the ControlNet scribble preprocessor to format the render into a scribble input. I took that scribble map and made some manual adjustments to suit what I wanted, and it worked fine - I think it was a time save compared to tracing everything myself. Here's an example of one of the scribble inputs I used, but I can't say if it's the one I used throughout or not because I had a few iterations on it.

I then got to genning a bunch of results (using a modified plasma noise that had been blue coloured) with different style/artist combinations that I thought could work well for the concept, and selected one to continue working on. I made liberal use of image editing and inpainting to touch areas up and add more details. The AI really didn't like having the back of an office chair, and kept trying to turn it into the woman's torso - it's likely down to it being an uncommon framing, and the hair being draped over the back probably didn't help the AI's interpretation.

I had bigger and messier dreams for the stack of pizza boxes, however ran into difficulties with getting anything successful from only/very lightly guided inpainting. Were I to return to this, I'd block something out in Blender or draw a good scribble to help the AI on its way. I'd also like to spruce the background up a bit more - there were other base results with some more interesting backgrounds, but worse foregrounds. I think the use of a basic scribble had some impact on the backgrounds, since those areas were blank in the scribble, the AI was dissuaded from adding anything there, except in the lower weights.

Very loosely inspired by etfffggfbut's suggestion .

These are the parameters for the original generation, but don't expect them to yield this result.
Prompt: solo, female, rear view, human, obese, (huge butt:1.3), fat rolls, thick thighs, (panties:1.2), long hair, BREAK sitting (office chair:1.2), office chair back, office desk, computer monitor lit screen, led light, blue glow, backlighting, detailed background, BREAK digital media \(artwork\), by hioshiru, by darkgem, by snowskau, by kenket, dark, blue monochrome
Negative prompt: greyscale, signature, multiple images, multiple scenes, back rolls
Other: Steps: 20, Sampler: Euler a, Cfg scale: 6, Seed: 85027719, Size: 640x512, Denoising strength: 0.87, Clip skip: 2, Controlnet 0 enabled: True, Controlnet 0 preprocessor: none, Controlnet 0 model: control_v11p_sd15_scribble [d4ba51ff], Controlnet 0 weight: 0.6, Controlnet 0 starting step: 0, Controlnet 0 ending step: 1, Controlnet 0 resize mode: Crop and Resize, Controlnet 0 pixel perfect: True, Controlnet 0 control mode: Balanced, Controlnet 0 preprocessor parameters: "(64, 64, 64)", Noise multiplier: 0.5
Model: Fluffusion Release 1 - Epoch 19 [3C8AECB4C2]

Description
It might not look it, but this was a fairly involved piece, with lots of editing and inpainting. This is one of those instances where I had a good idea of what I wanted to work towards and, as is usual with that, I eventually got frustrated when things weren't going as I envisioned, and gave up once the image was at an okayish state

As with a lot of ideas, I started off by blocking out a basic scene in Blender since I'm not great with visualising 3D spaces enough to just draw something freehand. Then, as a test to see how well it could work, I utilised the ControlNet scribble preprocessor to format the render into a scribble input. I took that scribble map and made some manual adjustments to suit what I wanted, and it worked fine - I think it was a time save compared to tracing everything myself. Here's an example of one of the scribble inputs I used, but I can't say if it's the one I used throughout or not because I had a few iterations on it.

I then got to genning a bunch of results (using a modified plasma noise that had been blue coloured) with different style/artist combinations that I thought could work well for the concept, and selected one to continue working on. I made liberal use of image editing and inpainting to touch areas up and add more details. The AI really didn't like having the back of an office chair, and kept trying to turn it into the woman's torso - it's likely down to it being an uncommon framing, and the hair being draped over the back probably didn't help the AI's interpretation.

I had bigger and messier dreams for the stack of pizza boxes, however ran into difficulties with getting anything successful from only/very lightly guided inpainting. Were I to return to this, I'd block something out in Blender or draw a good scribble to help the AI on its way. I'd also like to spruce the background up a bit more - there were other base results with some more interesting backgrounds, but worse foregrounds. I think the use of a basic scribble had some impact on the backgrounds, since those areas were blank in the scribble, the AI was dissuaded from adding anything there, except in the lower weights.

Very loosely inspired by etfffggfbut's suggestion .

These are the parameters for the original generation, but don't expect them to yield this result.
Prompt: solo, female, rear view, human, obese, (huge butt:1.3), fat rolls, thick thighs, (panties:1.2), long hair, BREAK sitting (office chair:1.2), office chair back, office desk, computer monitor lit screen, led light, blue glow, backlighting, detailed background, BREAK digital media \(artwork\), by hioshiru, by darkgem, by snowskau, by kenket, dark, blue monochrome
Negative prompt: greyscale, signature, multiple images, multiple scenes, back rolls
Other: Steps: 20, Sampler: Euler a, Cfg scale: 6, Seed: 85027719, Size: 640x512, Denoising strength: 0.87, Clip skip: 2, Controlnet 0 enabled: True, Controlnet 0 preprocessor: none, Controlnet 0 model: control_v11p_sd15_scribble [d4ba51ff], Controlnet 0 weight: 0.6, Controlnet 0 starting step: 0, Controlnet 0 ending step: 1, Controlnet 0 resize mode: Crop and Resize, Controlnet 0 pixel perfect: True, Controlnet 0 control mode: Balanced, Controlnet 0 preprocessor parameters: "(64, 64, 64)", Noise multiplier: 0.5
Model: Fluffusion Release 1 - Epoch 19 [3C8AECB4C2]