Generated on StableDiffusion with ControlNet. For ControlNet, I used the Red Dead Redemption 2 Poster, using depth_zoe preprocessor and the depth control model, “my prompt is more important”, Low VRAM enabled, and a control weight of 0.5. The generated image has some illegible text at the bottom, so I cropped that out. I also used a custom LORA of a certain Korean celebrity.
To get the prompt, I initially used the interrogator extension on Automatic1111 and changed it a little bit.
Prompt: a poster of (subject) holding a gun, in red dead redemption 2, 2d icon, seven stars in right hand, bbc promotional artwork, front page, images on the sales website, featured on ign, detailed face, digital art, digital painting,
Negative Embeddings: BadDream, UnrealisticDream, bad-hands-5
Model: DreamShaper
Sampler: DPM++ SDE Karras, 30 steps, upscaled with ESRGAN4x, 20 hires steps, 0.5 denoise
Honestly the gun isn’t that bad?
It would probably be ridiculous looking from any other angle, but from a pretty close to straight forward perspective… it’s fine.
I wanted to try SD myself, but from my understanding it’s really finicky with AMD GPUs so I’m SOL.