[TheWalkingDead1986] the stevensons remake with AI

thewalkingdead1986

New member
Hi guys, I decided to open this thread to upload my version of the Stevensons comic, a remake made with Ai and created with ComfyUI with my prompts, I hope you like it.

0.jpeg
 
thewalkingdead1986 said:
the first 27 page of the comic

View attachment 2137221 View attachment 2137222 View attachment 2137223 View attachment 2137224 View attachment 2137225 View attachment 2137226 View attachment 2137227 View attachment 2137228 View attachment 2137229 View attachment 2137230 View attachment 2137231 View attachment 2137232 View attachment 2137233 View attachment 2137234 View attachment 2137235 View attachment 2137236 View attachment 2137237 View attachment 2137238 View attachment 2137239 View attachment 2137240 View attachment 2137241 View attachment 2137242 View attachment 2137243 View attachment 2137244 View attachment 2137245 View attachment 2137246 View attachment 2137247 View attachment 2137248 View attachment 2137249 View attachment 2137250 View attachment 2137251 View attachment 2137252 View attachment 2137253 View attachment 2137254 View attachment 2137255 View attachment 2137256 View attachment 2137257 View attachment 2137258
Click to expand...
May I ask how you did this? What you used etc? really interesting, I think this is a brilliant concept. Might bring new life into older comics, I'd love to know how you did it, if you don't mind sharing.
 
Some AI generated images. Used Qwen Perchance AI image to image.
It's free but limited. Limited mages only per day (3 or 4). If someone is free please convert the other stevensons image as well. We can also add additional image for ref. So, I guess we can carry over the face of characters for next images. Not sure btw.

image.png image.png
 
Fantastic work! This AI tool could really update some of Dofantasy's old gems. Thanks for your contribution, and I hope to contribute someday.
 
Xolas888 said:
Fantastic work! This AI tool could really update some of Dofantasy's old gems. Thanks for your contribution, and I hope to contribute someday.
Click to expand...
It's free. But you have to be fast as you only get 2 mins per day. You can create minimum 2 images in 2 mins. You can connect to different wifi and get another 2 mins time as well.You can use the same prompt I used.
 
NarrowMinded said:
It's free. But you have to be fast as you only get 2 mins per day. You can create minimum 2 images in 2 mins. You can connect to different wifi and get another 2 mins time as well.You can use the same prompt I used.
Click to expand...
Which prompt should I use exactly to create the final images? I've done some tests, but they seem a bit blurry.
 
Xolas888 said:
Which prompt should I use exactly to create the final images? I've done some tests, but they seem a bit blurry.
Click to expand...
I already provided the prompt in my message. I type the prompt again. But texts won't be that clear sometimes. That's actually a trade off as the one I am using is free.

Prompt:-

Transform the cartoon image into photorealistic image. Don't remove any key details from this image like text. Retain all the texts in this image. Keep the women naked with big boobs.

Edited 7 images:-

image.png image (1).png image (2).png image.png image.png image (1).png image (2).png
 
Last edited:
NarrowMinded said:
Which website you used to convert this? What is the prompt? Is it free?
Click to expand...
I didn't use an online converter or a website for this — it’s a fully local workflow running on Stable Diffusion 1.5 via WebUI.

The whole pipeline breaks down into a few distinct steps:

  1. Vision-to-Prompt Analysis: I ran the original comic panel through Qwen-VL with a custom prompt. The goal was to deconstruct the image into real-world physical elements — describing lighting, materials, facial expressions, and clothing while completely stripping away any mention of comic styles, inking, or flat shading.
  2. Structural Composition: To lock the scene without copying the 2D illustration artifacts, I used ControlNet with a Depth map (depth_zoe). Depth retains the camera angle, spatial depth, and body poses, but avoids the rigid black outlines that Lineart or Canny usually bake into the render.
  3. Generation & Styling: The parsed prompt was paired with heavy photorealism modifiers (35mm photo, detailed skin texture, raw lighting) and a strict negative prompt to eliminate cel-shading, drawings, and artifacts.
  4. Face & Hands Refinement: Because SD 1.5 struggles with distant faces in multi-character scenes, I ran the output through ADetailer (face_yolov8s / hand_yolov8n) for automated high-res inpainting, ensuring clean expressions and anatomically correct hands.
So instead of a simple 2D-to-realism filter, it's essentially an automated reconstruction using a VLM for description and ControlNet for geometry.
 
Partyzan said:
I didn't use an online converter or a website for this — it’s a fully local workflow running on Stable Diffusion 1.5 via WebUI.

The whole pipeline breaks down into a few distinct steps:

  1. Vision-to-Prompt Analysis: I ran the original comic panel through Qwen-VL with a custom prompt. The goal was to deconstruct the image into real-world physical elements — describing lighting, materials, facial expressions, and clothing while completely stripping away any mention of comic styles, inking, or flat shading.
  2. Structural Composition: To lock the scene without copying the 2D illustration artifacts, I used ControlNet with a Depth map (depth_zoe). Depth retains the camera angle, spatial depth, and body poses, but avoids the rigid black outlines that Lineart or Canny usually bake into the render.
  3. Generation & Styling: The parsed prompt was paired with heavy photorealism modifiers (35mm photo, detailed skin texture, raw lighting) and a strict negative prompt to eliminate cel-shading, drawings, and artifacts.
  4. Face & Hands Refinement: Because SD 1.5 struggles with distant faces in multi-character scenes, I ran the output through ADetailer (face_yolov8s / hand_yolov8n) for automated high-res inpainting, ensuring clean expressions and anatomically correct hands.
So instead of a simple 2D-to-realism filter, it's essentially an automated reconstruction using a VLM for description and ControlNet for geometry.
Click to expand...
Thank you so much! As a Stable 1.5 user who is kinda stuck in a rut, this is extremely helpful.
 
You must log in or register to reply here.
Back
Top