
Create HYPERREALISTIC AI Characters That INTERACT | FREE & LOCAL
Build a whole cast of consistent AI characters, place them exactly where you want with bounding boxes, and put them in the same scene: all locally, with free open-weight tools.
If you like my work, please consider supporting me on Patreon: https://www.patreon.com/Mickmumpitz
Follow me on Twitter: https://twitter.com/mickmumpitz
The new generation of open image models is photorealistic enough now to rival, and in a few cases beat, closed tools like ChatGPT Image and Nano Banana Pro. This tutorial shows you the full pipeline to put them to work.
FREE WORKFLOWS: https://www.patreon.com/Mickmumpitz/posts/new-video-free-4-164075060
ComfyUI INSTALLATION GUIDE: https://mickmumpitz.ai/posts/comfyui-getting-153738117?l=de-DE
ADVANCED WORKFLOWS, EXAMPLE DATASETS & TRAINED LORAS:
Train your LoRAs on RunPod*: https://console.runpod.io/hub/template/ai-toolkit-ostris-ui-official?id=0fqzfjy6f3&ref=mgaruyku
What you'll learn:
Generate a consistent character dataset from a single reference image with Flux 2 Klein - Auto-caption it with a local vision model, then review and fix captions and bounding boxes - Train your entire cast into one LoRA in AI Toolkit, no bleeding between characters - Generate final images with Krea 2 and Ideogram 4, placing characters precisely with bounding boxes - Bring your cast into motion with Wan 2.2
TOOLS:
- ComfyUI
- Flux 2 Klein (Black Forest Labs)
- Krea 2 RAW / Turbo
- Ideogram 4
- Wan 2.2
- Qwen3-VL (captioning)
- SeedVR2 (upscaling)
- AI Toolkit by Ostris (LoRA training) — https://github.com/ostris/ai-toolkit
- Alternatives: Musubi Tuner (low VRAM / video) — https://github.com/kohya-ss/musubi-tuner · OneTrainer — https://github.com/Nerogar/OneTrainer - RES4LYF (samplers)
- RunPod (optional)
CHAPTERS:
00:00 Intro
00:48 How LoRAs actually work
03:05 Installation
03:48 Character Creator: realistic
09:00 Character Creator: stylized
09:44 Thank you! (+ Advanced version)
11:28 Captioning your dataset 12:30 Training the LoRA (+ RunPod) 15:50 Generating with Krea 2
16:55 Multiple characters
22:30 Bounding boxes and Ideogram 4
30:00 Video with Wan 2.2
*Affiliate link. If you sign up using our link, you'll get a random credit bonus between $5 and $500 when you spend your first $10 on the platform. Note for EU users: the bonus is a fixed $5 credit after your first $10 deposit.
If you like my work, please consider supporting me on Patreon: https://www.patreon.com/Mickmumpitz
Follow me on Twitter: https://twitter.com/mickmumpitz
The new generation of open image models is photorealistic enough now to rival, and in a few cases beat, closed tools like ChatGPT Image and Nano Banana Pro. This tutorial shows you the full pipeline to put them to work.
FREE WORKFLOWS: https://www.patreon.com/Mickmumpitz/posts/new-video-free-4-164075060
ComfyUI INSTALLATION GUIDE: https://mickmumpitz.ai/posts/comfyui-getting-153738117?l=de-DE
ADVANCED WORKFLOWS, EXAMPLE DATASETS & TRAINED LORAS:
Train your LoRAs on RunPod*: https://console.runpod.io/hub/template/ai-toolkit-ostris-ui-official?id=0fqzfjy6f3&ref=mgaruyku
What you'll learn:
Generate a consistent character dataset from a single reference image with Flux 2 Klein - Auto-caption it with a local vision model, then review and fix captions and bounding boxes - Train your entire cast into one LoRA in AI Toolkit, no bleeding between characters - Generate final images with Krea 2 and Ideogram 4, placing characters precisely with bounding boxes - Bring your cast into motion with Wan 2.2
TOOLS:
- ComfyUI
- Flux 2 Klein (Black Forest Labs)
- Krea 2 RAW / Turbo
- Ideogram 4
- Wan 2.2
- Qwen3-VL (captioning)
- SeedVR2 (upscaling)
- AI Toolkit by Ostris (LoRA training) — https://github.com/ostris/ai-toolkit
- Alternatives: Musubi Tuner (low VRAM / video) — https://github.com/kohya-ss/musubi-tuner · OneTrainer — https://github.com/Nerogar/OneTrainer - RES4LYF (samplers)
- RunPod (optional)
CHAPTERS:
00:00 Intro
00:48 How LoRAs actually work
03:05 Installation
03:48 Character Creator: realistic
09:00 Character Creator: stylized
09:44 Thank you! (+ Advanced version)
11:28 Captioning your dataset 12:30 Training the LoRA (+ RunPod) 15:50 Generating with Krea 2
16:55 Multiple characters
22:30 Bounding boxes and Ideogram 4
30:00 Video with Wan 2.2
*Affiliate link. If you sign up using our link, you'll get a random credit bonus between $5 and $500 when you spend your first $10 on the platform. Note for EU users: the bonus is a fixed $5 credit after your first $10 deposit.
Mickmumpitz
MickMumpitz is a channel that is all about understanding new technologies from the worlds of digital art, film and games in a playful and project-oriented way....