HiDream AI's 17B parameter text-to-image model using sparse diffusion transformer with mixture of experts, achieving state-of-the-art image generation quality with strong prompt following
Settled by community votes across 4 shared challenges, with an AI judge weighing in on each.
HiDream I1 Full
15.2
arena score
#60 of 62 in Text-to-Image
Skill signature
· Text-to-Image
P-Image Edit
9.5
arena score
#29 of 32 in Image Editing
Vote tally
Where the votes landed
HiDream I1 Full
0%
win rate
Ties
0%
P-Image Edit
0%
win rate
Shared challenges
4
Challenge by challenge
The strongest take from each model on every shared challenge, with the AI judge's read.
Bald man challenge
Image EditingEdit instruction
“Give the person a full, thick head of natural hair with realistic texture, density, and a natural hairline. Preserve facial features and lighting.”
HiDream I1 Full
P-Image Edit
AI Judge Analysis
HiDream I1 Full
- + Attempts to add hair length behind the head.
- − Failed to preserve the source image identity, background, and lighting.
- − The subject remains mostly bald on top, failing the primary prompt.
- − Introduced major artifacts like a strange mechanical device floating on the face.
P-Image Edit
- + Successfully added a full, thick head of hair as requested.
- + Perfectly preserved the subject's facial features and the original background.
- + Maintained consistent lighting and texture between the new hair and the rest of the image.
- − The hairline is a bit high and rounded.
- − The hair texture is slightly more uniform/stylized than 'natural'.
Verdict: P-Image Edit followed the instructions perfectly by adding a full head of hair while keeping the person and background exactly the same as the source image. HiDream I1 Full failed the task entirely, generating a different person in a different environment with bizarre visual artifacts and very little hair on top.
Over-the-top cartoon caricature
EditingEdit instruction
“Create a caricature of me and my job. Make it exaggerated and humorous, incorporating my profession as a tv show anchor and my love for dogs and hockey.”
Source
HiDream I1 Full
P-Image Edit
AI Judge Analysis
HiDream I1 Full
- + Successfully captures a more accurate facial likeness from the source image.
- + Includes the dog and denim jacket clearly.
- − Fails to incorporate any hockey-related elements.
- − The TV anchor profession is only vaguely hinted at by a screen in the background.
P-Image Edit
- + Fully incorporates all prompt elements including hockey players, a TV studio, and a dog.
- + The composition is highly creative and structured as a full scene.
- − The facial likeness is less accurate, appearing more like a generic cartoon character.
- − The text on the screens and signs is garbled and nonsensical.
Verdict: P-Image Edit (Model B) followed the complex editing instructions much more effectively by incorporating the hockey and TV anchor themes, whereas HiDream I1 Full (Model A) largely ignored the hockey request. While HiDream I1 Full preserved the subject's face better, P-Image Edit created a far more complete and humorous caricature as requested.
Studio Ghibli Anime Style
EditingEdit instruction
“Transform this photo into a Studio Ghibli–inspired illustration. Use soft pastel colors, hand-painted textures, gentle lighting, dreamy backgrounds, and a warm, nostalgic mood”
Source
HiDream I1 Full
P-Image Edit
AI Judge Analysis
HiDream I1 Full
- + Excellently captures the specific Studio Ghibli art style including line work and shading
- + Creative reinterpretation of the scene with added nostalgic elements like the hat and floral patterns
- − Lower source preservation as it changes the characters' identities and clothing patterns significantly
- − The 'distracted' narrative of the original meme is weakened by the changed expressions
P-Image Edit
- + High source preservation, maintaining the exact poses and facial features of the original people
- + Successfully applies a hand-painted texture and soft pastel color palette
- − The art style is more generic 'webtoon' or 'manhua' than specifically Ghibli-inspired
- − Lack of the distinctive Ghibli line art and character design philosophy
Verdict: HiDream I1 Full perfectly captures the requested Studio Ghibli aesthetic but does so by essentially redrawing the image into a new scene, losing the likeness of the original subjects. P-Image Edit preserves the original 'Distracted Boyfriend' meme perfectly and applies a soft painted texture, but fails to replicate the iconic Ghibli art style, resulting in a more generic digital illustration look. For an image editing task where the Ghibli style was the primary request, HiDream I1 Full is more successful despite the loss of source identity.
Golden Hour Stroll
Image EditingEdit instruction
“Add dynamic motion to this photo: make hair blow in the wind, add leaves flying, energetic and lively feel.”
HiDream I1 Full
P-Image Edit
AI Judge Analysis
HiDream I1 Full
- + Excellent sense of motion and energy
- + Effective hair blowing effect
- + Abundance of autumn leaves creates a lively atmosphere
- − Fails as an edit by completely regenerating the subject face and removing the dog
- − Alters the background significantly
- − Loss of source image identity
P-Image Edit
- + Perfect source preservation of the subject's face, clothing, and the dog
- + Accurately applies blowy hair effect while keeping the person recognizable
- + Integrates flying leaves without compromising original composition
- − The 'wind' effect on the hair is slightly symmetrical and looks a bit digital
- − Leaves are somewhat small and less impactful than Model A
Verdict: P-Image Edit is the clear winner for an image editing task as it successfully follows the instructions while perfectly preserving the identity of the woman and the presence of the dog. HiDream I1 Full failed to preserve the source image, effectively generating a completely new image of a different person and removing the dog entirely.
Next steps
Explore each model
PrunaAI's sub-1-second multi-image editing model supporting up to 5 reference images with state-of-the-art quality