I use a bunch of different tools to create images and videos.
For image generation, I start by making a base image/draft. I either draw it in Photoshop, do a photobash from several images from the internet, or use LLM image generators like ChatGPT or Qwen. If the image is complex and I expect to animate it or make variations, I generate the background separately first and only then add the characters.
After I have the initial draft, I run it through img2img locally (Stable Diffusion — I use A1111). I use several models, mostly from the Ilustrious family and the Pony family. Most of the time I only use LoRAs for the characters. Occasionally I add LoRAs for concepts the model doesn’t understand or keeps mixing up and needs a bit of fine-tuning. You have to be careful with that though, because LoRAs often shift the style — and that hurts consistency.
Then I take the result into Photoshop, fix what I don’t like, and send it back into Stable Diffusion for inpainting. I repeat that loop until I’ve fixed everything that bothers me.
Once I have a finished image, I can move on to video generation. I use WAN 2.2 models — there’s a pretty good variant by dasiwa. The “clean” model keeps consistency better, but then you end up stacking at least 5–6 LoRAs, and that affects face consistency.
I generate videos at 480p, tweaking the prompt until I get a version I like in terms of motion and/or it contains the frames I need. Then I generate a couple of variations with the prompt/seed I liked at 864p. After that, I pull out a frame I like, upscale it to 1080p, run it through img2img, and fix everything I don’t like. Color needs special attention — WAN “flattens” and desaturates colors.
Then I run the video in start-to-end frames mode once I have both frames ready.
Finally, I assemble the finished videos in DaVinci Resolve and convert them to WebM using Shutter Encoder.
That’s the rough flow — but there are nuances at every step
The 2-3 frames animation are the "old" ones I made by hand. The "new" AI animations look like this
It's a 16 fps base video that enhanced to 48/60 fps. Most of them 12 sec long. Some more then 30 sec.
Also, it's a bit of surprise for me, but looks like I will be switching to the loyal route after the NY holidays. Though it's not set in stone, people are still voting
View attachment 5581738