Testing 9 image models side-by-side

When I need a model for a specific task, I usually start on OpenRouter to see what other people are actually using. That usually gives me a shortlist. Finding the best model for the job involves some benchmarking and evaluation of my own. This time I need a model capable of generating fast, cheap and quality images.

The OpenRouter rankings narrowed a field of dozens to nine: two Riverflow, two Krea, three Gemini, Seedream and Grok Imagine.

I started each prompt with the following:

ultra realistic photograph, cinematic lighting, 35mm, high detail, vertical composition

Then one of six retail scenes, each picked to be awkward in a different way:

1. Hard product on seamless: the staple ecommerce shot. Tests a believable contact shadow under a floating object, a clean white plate, and separation between two materials.

2. Apparel on a model: a human figure. Tests anatomy, hands, and how fabric drapes, which is where image models most often fall apart.

3. Cosmetics and liquid: refraction. Tests frosted glass, wet surfaces, water droplets and steam, all of which need consistent light to look real.

4. Overhead flat lay: a camera instruction. Tests whether the model holds a true top-down angle instead of drifting to the three-quarter view it tends to produce.

5. Macro metal and gemstone: micro-detail. Tests specular highlights, prismatic sparkle and shallow depth of field on a small reflective object.

6. Retail environment: repeated geometry and a negative instruction. Tests perspective down a row of shelving, and whether “no people” is obeyed.

Every model got the identical prompt and a requested 9:16 portrait. Total cost 54 images from 9 models $2.92.

The results

Model $/image 6-scene set Avg latency Native size
sourceful/riverflow-v2.5-fast $0.016 $0.10 23.4s 576×1024
krea/krea-2-medium $0.030 $0.18 15.3s 768×1376
google/gemini-3.1-flash-lite-image $0.034 $0.20 3.9s 768×1376
google/gemini-2.5-flash-image $0.039 $0.23 8.5s 768×1344
bytedance-seed/seedream-4.5 $0.040 $0.24 15.9s 2048×3642
x-ai/grok-imagine-image-quality $0.050 $0.30 6.5s 720×1280
krea/krea-2-large $0.060 $0.36 32.8s 768×1376
google/gemini-3.1-flash-image $0.068 $0.41 9.7s 768×1376
sourceful/riverflow-v2-pro $0.150 $0.90 48.4s 720×1280

Scroll sideways; click any image to see it full size. Native pixel size is under each tile.

riverflow-v2.5-fastkrea-2-mediumgemini-3.1-flash-lite-imagegemini-2.5-flash-imageseedream-4.5grok-imagine-image-qualitykrea-2-largegemini-3.1-flash-imageriverflow-v2-pro
1. Hard product on seamlessultra realistic photograph, cinematic lighting, 35mm, high detail, vertical composition, a single premium white and grey running sneaker floating above a seamless white studio backdrop with a soft contact shadow beneath it, clean softbox product lighting, crisp material detail in the knit upper and rubber sole
Hard product on seamless rendered by riverflow-v2.5-fast576×1024Hard product on seamless rendered by krea-2-medium768×1376Hard product on seamless rendered by gemini-3.1-flash-lite-image768×1376Hard product on seamless rendered by gemini-2.5-flash-image768×1344Hard product on seamless rendered by seedream-4.52048×3642Hard product on seamless rendered by grok-imagine-image-quality720×1280Hard product on seamless rendered by krea-2-large768×1376Hard product on seamless rendered by gemini-3.1-flash-image768×1376Hard product on seamless rendered by riverflow-v2-pro720×1280
2. Apparel on a modelultra realistic photograph, cinematic lighting, 35mm, high detail, vertical composition, a young woman wearing an oversized beige chunky knit sweater and straight-leg jeans standing against a warm neutral studio backdrop, natural window light from the left, relaxed fashion lookbook pose, sharp fabric texture
Apparel on a model rendered by riverflow-v2.5-fast576×1024Apparel on a model rendered by krea-2-medium768×1376Apparel on a model rendered by gemini-3.1-flash-lite-image768×1376Apparel on a model rendered by gemini-2.5-flash-image768×1344Apparel on a model rendered by seedream-4.52048×3642Apparel on a model rendered by grok-imagine-image-quality720×1280Apparel on a model rendered by krea-2-large768×1376Apparel on a model rendered by gemini-3.1-flash-image768×1376Apparel on a model rendered by riverflow-v2-pro720×1280
3. Cosmetics and liquidultra realistic photograph, cinematic lighting, 35mm, high detail, vertical composition, an unbranded matte frosted-glass serum bottle with a dropper cap standing on a wet dark stone slab, scattered water droplets and gentle steam, soft directional morning light, minimal spa aesthetic
Cosmetics and liquid rendered by riverflow-v2.5-fast576×1024Cosmetics and liquid rendered by krea-2-medium768×1376Cosmetics and liquid rendered by gemini-3.1-flash-lite-image768×1376Cosmetics and liquid rendered by gemini-2.5-flash-image768×1344Cosmetics and liquid rendered by seedream-4.52048×3642Cosmetics and liquid rendered by grok-imagine-image-quality720×1280Cosmetics and liquid rendered by krea-2-large768×1376Cosmetics and liquid rendered by gemini-3.1-flash-image768×1376Cosmetics and liquid rendered by riverflow-v2-pro720×1280
4. Overhead flat layultra realistic photograph, cinematic lighting, 35mm, high detail, vertical composition, overhead flat lay of scattered roasted coffee beans, a matte ceramic mug of black coffee and a folded kraft paper pouch arranged on a dark walnut table, warm directional light casting soft shadows
Overhead flat lay rendered by riverflow-v2.5-fast576×1024Overhead flat lay rendered by krea-2-medium768×1376Overhead flat lay rendered by gemini-3.1-flash-lite-image768×1376Overhead flat lay rendered by gemini-2.5-flash-image768×1344Overhead flat lay rendered by seedream-4.52048×3642Overhead flat lay rendered by grok-imagine-image-quality720×1280Overhead flat lay rendered by krea-2-large768×1376Overhead flat lay rendered by gemini-3.1-flash-image768×1376Overhead flat lay rendered by riverflow-v2-pro720×1280
5. Macro metal and gemstoneultra realistic photograph, cinematic lighting, 35mm, high detail, vertical composition, extreme close-up of a polished yellow gold solitaire ring with a brilliant-cut diamond resting on black velvet, razor-sharp specular highlights and prismatic sparkle, jewellery catalogue lighting, shallow depth of field
Macro metal and gemstone rendered by riverflow-v2.5-fast576×1024Macro metal and gemstone rendered by krea-2-medium768×1376Macro metal and gemstone rendered by gemini-3.1-flash-lite-image768×1376Macro metal and gemstone rendered by gemini-2.5-flash-image768×1344Macro metal and gemstone rendered by seedream-4.52048×3642Macro metal and gemstone rendered by grok-imagine-image-quality720×1280Macro metal and gemstone rendered by krea-2-large768×1376Macro metal and gemstone rendered by gemini-3.1-flash-image768×1376Macro metal and gemstone rendered by riverflow-v2-pro720×1280
6. Retail environmentultra realistic photograph, cinematic lighting, 35mm, high detail, vertical composition, interior of a modern minimalist boutique clothing store with warm oak shelving, neatly folded garments in muted tones and a rail of hanging coats, soft overhead spotlights and polished concrete floor, no people
Retail environment rendered by riverflow-v2.5-fast576×1024Retail environment rendered by krea-2-medium768×1376Retail environment rendered by gemini-3.1-flash-lite-image768×1376Retail environment rendered by gemini-2.5-flash-image768×1344Retail environment rendered by seedream-4.52048×3642Retail environment rendered by grok-imagine-image-quality720×1280Retail environment rendered by krea-2-large768×1376Retail environment rendered by gemini-3.1-flash-image768×1376Retail environment rendered by riverflow-v2-pro720×1280

So far I think the best option is google/gemini-3.1-flash-lite-image, on price, speed and quality together. I have not settled on it yet. There are more benchmarks to run and more prompts to try before one of these goes into the pipeline.

What do you think? If you have run something similar, or you would have picked a different model off that grid, I would like to hear about it.