Each capability below is a fixed experiment: the same prompt and the same input images
for every model we test. That's what makes these outputs comparable — and it's why the
same bicycle shows up on every image model's page.
Each of these is a fixed experiment — the same prompt and the same input images
for every model we test, so the outputs are directly comparable.
Honors requested aspect ratio✓ passed
We asked
“Generate a wide 16:9 landscape photo of a mountain range.”
We gave it
Nothing — prompt only
It returned
the output’s dimensions were measured5.3s
Renders a data-faithful infographic✓ passed
We asked
“Create a clean infographic: a vertical bar chart titled 'HIFIBOTS USERS' with exactly three labeled bars: Jan = 10, Feb = 25, Mar = 40. Show each label and value as legible text.”
We gave it
Nothing — prompt only
It returned
a vision model was asked a yes/no question about the output4.1s
Replaces a target object in place (prompt-based inpainting)✓ passed
We asked
“In this image, replace the red square with a yellow star. Keep the blue circle exactly as it is.”
We gave it 1 reference image
shapes.png
It returned
a vision model was asked a yes/no question about the output15s
Blends multiple reference images✓ passed
We asked
“Combine these two images into one scene keeping the red square and the blue circle.”
We gave it 2 reference images
shapes.pngcount.png
It returned
a vision model was asked a yes/no question about the output5.6s
Accepts reference image (I2I edit)✓ passed
We asked
“Edit this image: change the red square to green. Keep everything else identical.”
We gave it 1 reference image
shapes.png
It returned
passed if the model returned an image4.2s
Applies a style to a reference✓ passed
We asked
“Redraw this image in the style of a watercolor painting, keeping the shapes.”
We gave it 1 reference image
shapes.png
It returned
passed if the model returned an image4.6s
Renders legible text in image✓ passed
We asked
“Generate an image of a wooden sign with the word HIFIBOTS painted clearly on it.”
We gave it
Nothing — prompt only
It returned
a vision model was asked a yes/no question about the output5.1s
Text to image✓ passed
We asked
“A red bicycle leaning against a green wall, photorealistic”
We gave it
Nothing — prompt only
It returned
passed if the model returned an image5.0s
Upscales / super-resolution✓ passed
We asked
“Upscale this image to a higher resolution, keeping content identical.”
We gave it 1 reference image
shapes.png
It returned
“ratio=1.82 want=1.82”
the output’s dimensions were measured3.6s
Removes background✗ did not pass
We asked
“Remove the background from this image, leaving only the shapes on transparency.”
We gave it 1 reference image
shapes.png
It returned
the returned pixels were inspected for a real alpha channel6.2s
Outputs transparent background (alpha)✗ did not pass
We asked
“Generate a single red circle on a fully transparent background. Output a PNG with alpha.”
We gave it
Nothing — prompt only
It returned
the returned pixels were inspected for a real alpha channel6.3s
Specs
latency s
5.3040735721588135
speed tier
fast
Where to run it
The same model, priced by each provider that serves it. Cheapest first — the card's
headline price is the best available anywhere, which is the wrong number for choosing
where to actually send your traffic.
Every claim is tagged by how we know it. Tested means we ran the model against a fixed
prompt and checked the output — the images above are those runs. Researched means it's
documented by the provider and cited. Nothing here is inferred.
