This is an AI image generation comparison for
reference-to-image 
prompt:
A woman in 1940s workwear sits on the edge of a steel I-beam high above a sprawling city. She wears a light blue denim shirt with rolled-up sleeves, gray work pants, lace-up boots, white gloves, and a red polka-dot bandana. Seated in a three-quarter view, her body faces left while her gaze looks ahead, giving her a thoughtful, stoic air. Her feet rest firmly on the beam, arms on her thighs, exuding quiet confidence. Positioned in the lower third of the frame, the beam stretches beneath her, with...
Log in to see full prompt.
Tested: May 18, 2025
Reference to image using Hailuo's Image-01 model.
Tested: May 18, 2025
KOLORS 1.5 model done pretty well here.
Tested: May 18, 2025
Best of 4 generations of Seedream 2.1 using reference face.
Tested: May 18, 2025
Tested: May 18, 2025
Using ByteDance/DreamO spaces
Tested: July 3, 2025
Tested: July 3, 2025
Longer more detailed prompt loses reference face entirely.
Tested: July 30, 2025
Tested: August 1, 2025
This is 1/8 genereations and it's hardly usable. Seems like Photon image reference model is underbaked.
Tested: August 27, 2025
So this same ref image works in Gemini but gets banned in Whisk, and doesn't seem to produce precise outcomes in Freepik.
Can you recognize the woman as the girl with pearl earring formt the famous painting? Is she seated on a steel I-beam positioned high above a visible cityscape?
Is she shown in three-quarter view, with her body turned left and her gaze directed forward?
Is she wearing a light blue denim shirt with rolled-up sleeves?
Are her pants gray work trousers, not form-fitting or modern in style?
Are lace-up boots visible and appropriate for a 1940s worker?
Is she wearing white gloves and a red polka-dot bandana?
Are her arms resting on her thighs, with a calm, confident posture?
Is the beam extending across the lower third of the frame with her seated on it?
Check out the results from Hailuo AI x Minimax (Hailuo Image-01) vs Kling AI vs Dreamina AI (Seedream 2.1) vs Pulid vs DreamO vs Runway (Runway's Gen-4 Image) vs FLUX.1 Kontext (FLUX Kontext [Max]) vs Ideogram (Ideogram Character) vs Luma Dream Machine vs Google Gemini App (Google Nano Banana) for similar or identical prompts side-by-side.
