Black Forest Labs' compact, open-source image generation model with sub-second inference, optimized for production and near real-time applications with multi-reference support
Settled by community votes across 4 shared challenges, with an AI judge weighing in on each.
FLUX.2 [klein] 4B
22.3
arena score
#32 of 62 in Text-to-Image
Skill signature
· Text-to-Image
HiDream E1
0.7
arena score
#33 of 32 in Image Editing
Vote tally
Where the votes landed
FLUX.2 [klein] 4B
0%
win rate
Ties
0%
HiDream E1
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.”
FLUX.2 [klein] 4B
HiDream E1
AI Judge Analysis
FLUX.2 [klein] 4B
- + Excellent preservation of original facial features and expression
- + Highly realistic hair texture and natural integration around the ears
- + Maintains the original background and clothing lighting perfectly
- − None notable
HiDream E1
- + Successfully adds a full head of hair
- − Heavily alters the background to a desert dune scene
- − Significant degradation of facial textures, making the skin look blotchy
- − Hair appears overly smoothed and artificial compared to the requested realistic texture
Verdict: FLUX.2 [klein] 4B follows the edit instructions perfectly, adding realistic hair while completely preserving the existing face, clothing, and background. In contrast, HiDream E1 fails as an editing tool by unnecessarily regenerating the background and altering the subject's facial details, resulting in a lower-quality overall image.
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
FLUX.2 [klein] 4B
HiDream E1
AI Judge Analysis
FLUX.2 [klein] 4B
- + Captures the 'TV show anchor' setting perfectly with the news desk and studio background.
- + Maintains the likeness of the source person through the caricature style.
- + Includes two well-illustrated dogs as requested.
- − Completely missed the 'hockey' requirement in the final image.
- − The background screens contain illegible and distorted text/graphics.
HiDream E1
- + Successfully incorporates all three elements: dogs, TV personality, and hockey players.
- + High-energy caricature style with exaggerated facial features.
- + Clear and vibrant visual quality.
- − The character's likeness has drifted quite far from the original source image.
- − Significant anatomical issues with the hands and the number of legs on the dog.
Verdict: HiDream E1 was more successful at incorporating all of the requested prompt elements, including the hockey requirement which FLUX.2 [klein] 4B missed. However, FLUX.2 [klein] 4B followed the source image's likeness much more accurately and produced a cleaner overall composition. HiDream E1 suffers from notable AI artifacts in the dog's legs and the subject's hands, making FLUX.2 [klein] 4B the better technical choice despite the missing hockey element.
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
FLUX.2 [klein] 4B
HiDream E1
AI Judge Analysis
FLUX.2 [klein] 4B
- + Excellent preservation of the original subjects' poses and clothing details.
- + Highly accurate Studio Ghibli watercolor texture and soft pastel palette.
- + Superior preservation of the background architecture and composition.
- − The facial expressions are significantly softened, making the 'jealous' partner look happy rather than angry.
HiDream E1
- + Successfully translates the scene into a flat 2D anime style.
- + Maintains the general character positioning of the source image.
- − Loses the specific texture and nostalgic feel of the Ghibli style in favor of generic modern anime aesthetic.
- − Simplified the man's shirt to a solid color, losing the original plaid detail.
- − The background is very blurry and lacks the detailed 'dreamy' hand-painted look requested.
Verdict: FLUX.2 [klein] 4B is the clear winner as it masterfully balances the Ghibli watercolor aesthetic with the specific details of the source image, such as the man's plaid shirt and the urban background. HiDream E1 produces a more generic, flat anime style that loses many of the original photo's defining features and fails to capture the 'hand-painted' texture requested.
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.”
FLUX.2 [klein] 4B
HiDream E1
AI Judge Analysis
FLUX.2 [klein] 4B
- + Excellent preservation of the original subjects' features and faces.
- + The flying leaves match the species and lighting of the background trees.
- + The hair motion is subtle and looks more physically realistic.
- − The leaf placement looks slightly cluttered across the center of the image.
HiDream E1
- + Successfully adds a dramatic sense of motion through pronounced hair flow.
- + Bold, high-contrast leaves create a clear visual indicator of the requested edit.
- − The leaves appear like 2D stickers and do not match the park environment's foliage.
- − The woman's facial features were altered significantly compared to the source image.
- − Overall image quality looks over-processed and loses some fine detail.
Verdict: FLUX.2 [klein] 4B is the clear winner because it successfully integrates the requested edits—wind-blown hair and flying leaves—while perfectly preserving the identity and lighting of the source image. HiDream E1 makes the hair more dynamic, but the added leaves look artificial and the model fails to preserve the subject's face accurately.
Next steps
Explore each model
HiDream AI's image-to-image editing model for instruction-based image modifications and transformations