(white goat:1.2), angry smirk, (short blue horns:1.2), pointed snout, (purple beard:1.05), stylized, ruins of an ancient city, detailed background, glowing blue eyes, ragemode, <lora:rageMode_v1:0.5> Negative prompt: lowres, poorly drawn, out of frame, (cropped), ((worst quality)), ((low quality)), jpeg artifacts, text, error, signature, watermark, logo, username, artist name, blurry, out of focus Steps: 24, Sampler: DDIM, CFG scale: 9, Seed: 948454521, Size: 640x360, Model hash: 6e430eb514, Model: anythingelseV4_v45, Clip skip: 2, Version: v1.2.1, ControlNet: "preprocessor: lineart_anime_denoise, model: control_v11p_sd15_lineart [43d4be0d], weight: 1, starting/ending: (0, 1), resize mode: Just Resize, pixel perfect: False, control mode: Balanced, preprocessor params: (360, 0.4, 0.1)"
The RageMode LoRA at 0.5 strength provides apocalyptic atmosphere and empowered goat presence. The kl_f8_anime2 VAE was applied to the AnythingElse v4.5 model for the stylized comic book look. ControlNet forces the classic voatgoat shape.
First the voatgoat.png linked in OP was resized in a separate image tool. Then preprocessed with the lineart_anime_denoise model of ControlNet, to produce this template (you could just reuse it with 'none' as preprocessor):
Then batches of 640x360 samples were generated as shown above, to find promising seeds. Good seeds were reused to regen with HiRes fix applying Latent x2, typically with 0.6 denoise, producing 1280x720 intermediate images.
If these looked good as well, they were sent to img2img. There, they were resized x1.5 with DDIM x24 @ 0.4-0.5 denoise, producing the final 1920x1080 resolution. The ragemode lora was removed from the prompt at this stage, to get a clean upscale without distorting the scene further. Sometimes additional terms were added to the prompt instead, to steer the finer details in the right direction. ControlNet was no longer needed at this point.
Can't make any promises, especially when working for free, but I'd probably give it a shot to help a goat out if you tell me what you have in mind. I've never tried making a logo or anything of that nature though. This is pretty much the opposite of that. And I'm pretty close to useless at traditional image editing. Just to keep your expectations in check.
[ + ] SilentByAssociation
[ - ] SilentByAssociation 1 point 1.9 yearsMay 29, 2023 13:55:48 ago (+1/-0)
[ + ] GloryBeckons
[ - ] GloryBeckons [op] 0 points 1.9 yearsMay 29, 2023 10:52:34 ago (+0/-0)
https://files.catbox.moe/m9cgsu.png
https://files.catbox.moe/6r2lv2.png
https://files.catbox.moe/17yqwx.png
https://files.catbox.moe/v4oshn.png
https://files.catbox.moe/dpyueo.png
https://files.catbox.moe/vuwjsg.png
Mirrors:
https://pomf2.lain.la/f/zjvgm4t3.png
https://pomf2.lain.la/f/s9rei7h.png
https://pomf2.lain.la/f/u1yf9nl3.png
https://pomf2.lain.la/f/k0ban77d.png
https://pomf2.lain.la/f/55p2vqka.png
https://pomf2.lain.la/f/624xrp5w.png
[ + ] GloryBeckons
[ - ] GloryBeckons [op] 0 points 1.9 yearsMay 29, 2023 11:29:21 ago (+0/-0)
Sample prompt info:
Negative prompt: lowres, poorly drawn, out of frame, (cropped), ((worst quality)), ((low quality)), jpeg artifacts, text, error, signature, watermark, logo, username, artist name, blurry, out of focus
Steps: 24, Sampler: DDIM, CFG scale: 9, Seed: 948454521, Size: 640x360, Model hash: 6e430eb514, Model: anythingelseV4_v45, Clip skip: 2, Version: v1.2.1, ControlNet: "preprocessor: lineart_anime_denoise, model: control_v11p_sd15_lineart [43d4be0d], weight: 1, starting/ending: (0, 1), resize mode: Just Resize, pixel perfect: False, control mode: Balanced, preprocessor params: (360, 0.4, 0.1)"
The RageMode LoRA at 0.5 strength provides apocalyptic atmosphere and empowered goat presence. The kl_f8_anime2 VAE was applied to the AnythingElse v4.5 model for the stylized comic book look. ControlNet forces the classic voatgoat shape.
First the voatgoat.png linked in OP was resized in a separate image tool. Then preprocessed with the lineart_anime_denoise model of ControlNet, to produce this template (you could just reuse it with 'none' as preprocessor):
https://files.catbox.moe/tib6vm.png
https://pomf2.lain.la/f/8mbh9a9g.png
Then batches of 640x360 samples were generated as shown above, to find promising seeds. Good seeds were reused to regen with HiRes fix applying Latent x2, typically with 0.6 denoise, producing 1280x720 intermediate images.
If these looked good as well, they were sent to img2img. There, they were resized x1.5 with DDIM x24 @ 0.4-0.5 denoise, producing the final 1920x1080 resolution. The ragemode lora was removed from the prompt at this stage, to get a clean upscale without distorting the scene further. Sometimes additional terms were added to the prompt instead, to steer the finer details in the right direction. ControlNet was no longer needed at this point.
[ + ] v0atmage
[ - ] v0atmage 0 points 1.9 yearsMay 29, 2023 13:08:59 ago (+0/-0)
[ + ] GloryBeckons
[ - ] GloryBeckons [op] 0 points 1.9 yearsMay 29, 2023 13:28:00 ago (+0/-0)
[ + ] v0atmage
[ - ] v0atmage 1 point 1.9 yearsMay 29, 2023 15:30:09 ago (+1/-0)