I recently developed a workflow that lets me generate photorealistic images with a level of quality I haven’t seen before in person-specific AI generation.
Yes, LoRA trainers can already keep a person consistent across different scenes—but the quality has always been the real bottleneck for my customers. I think I’ve found a way to balance facial consistency while allowing for natural variation across completely different contexts.
After months of experimenting in ComfyUI, I've built a workflow that can take a few photos of someone and generate them convincingly with natural lighting, authentic expressions, and true photorealism.
Here are some examples of what the workflow can produce of the same person:
The real breakthrough came from completely rethinking how LoRAs handles lightning and expression. I focused on fixing the core quality issues that make AI photos look.. well, like AI photos.
We're currently working on making these high-quality images available to Premium and Ultra Users of MyPicNow.com, where people can create professional-grade photos of themselves from just a few reference shots.
What's been your experience with person-specific image generation that actually generates such high-quality images? I'm using Flux as the base, but curious if others have found ways to push realism to this level.
mypicnow•6h ago
I recently developed a workflow that lets me generate photorealistic images with a level of quality I haven’t seen before in person-specific AI generation.
Yes, LoRA trainers can already keep a person consistent across different scenes—but the quality has always been the real bottleneck for my customers. I think I’ve found a way to balance facial consistency while allowing for natural variation across completely different contexts.
After months of experimenting in ComfyUI, I've built a workflow that can take a few photos of someone and generate them convincingly with natural lighting, authentic expressions, and true photorealism.
Here are some examples of what the workflow can produce of the same person:
- Tourist photo at the Eiffel tower in Paris -> https://imgur.com/a/OxZNklm
- Professional speaker at a TedX event -> https://imgur.com/a/KOC6eCy
- Rockstar performer with dramatic concert lighting -> https://imgur.com/a/B2YaUHj
- Casual sunset portrait on a boat -> https://imgur.com/a/d4rEANW
The real breakthrough came from completely rethinking how LoRAs handles lightning and expression. I focused on fixing the core quality issues that make AI photos look.. well, like AI photos.
We're currently working on making these high-quality images available to Premium and Ultra Users of MyPicNow.com, where people can create professional-grade photos of themselves from just a few reference shots.
What's been your experience with person-specific image generation that actually generates such high-quality images? I'm using Flux as the base, but curious if others have found ways to push realism to this level.