Two women dancing under a disco ball — GPT Image 2 render.
Friends of Fal A new daily driver?

GPT Image 2
is here.

Cheap. Coherent. Sharp on type. A real contender to replace Nano Banana Pro as the default.

On fal.ai — text-to-image + mask edit. Commercial via fal Partner.

Coherent Text rendering Photorealism Mask edit Up to 4K Commercial
01 · What's new

Why it's my
new daily driver.

01

Richer lighting and framing

Less common-denominator training. More dynamic angles and compositions across everything.

02

Photorealistic, cinematic output

Physics, lighting, materials. The warm cast is gone.

03

Phenomenal prompt adherence

Holds composition across long, multi-part briefs.

04

Near-perfect text rendering

Signage, UI, posters. Latin and CJK.

05

Custom sizes up to 4K

Edges in multiples of 16, 3840px max, 3:1 aspect max.

One caveat There are still some noticeable artefacts/compression style faint markings - similar to the same texture that appears when nano banana images are run over and over again. I'll be investigating, tbc. 🔍
Possum ballerinas on red — GPT Image 2.
02 · Side by side

GPT vs NBP.

Same prompt, two models. Toggle below, use ← → to flick between scenes.

Fig. 01 · 10 scenes · same prompt, both models Click thumbnail · ← → to navigate
03 · Quality tiers

Low, medium, high.

Same prompt, three tiers. What the extra cents buy.

Macro grapes — low quality.
Macro grapes — medium quality.
Macro grapes — high quality.

Prompt

Macro close-up of jewel-toned grapes on hot-pink velvet. Shallow DoF, 100mm macro at f/4, editorial food photography.
04 · Prompt discipline

Vague in,
vague out.

Same concept, three lengths. Watch it sharpen.

8 words

Billboard in Sydney advertising a new coffee brand.

Short-prompt billboard render.
05 · Where it fits

The new default
for four jobs.

01

Realistic photography

Cinematic, editorial, documentary realism.

02

Any job where adherence matters

Long prompts, detailed scenes.

03

Exploration of stylistic type

Headlines, wordmarks, expressive set type.

04

Mockups with legible text

Packaging, posters, OOH, UI.

06 · Range

24 hours of testing.

Pop-art collage moodboard. Click to copy prompt
Editorial
UV lip marks on fabric. Click to copy prompt
Editorial
Candy apple on beach. Click to copy prompt
Product
Slushie packaging mockup. Click to copy prompt
Packaging
Photo booth grid portrait. Click to copy prompt
Editorial
Arcade neon portrait. Click to copy prompt
Narrative
Narrative scene — woman in toy room. Click to copy prompt
Narrative
Fashion blue shorts detail. Click to copy prompt
Product
Girl in bedroom with books. Click to copy prompt
Narrative
B&W cat contact sheet pinned to cork board. Click to copy prompt
Editorial
Hotel pool at sunset. Click to copy prompt
Editorial
07 · What stays

Not for everything.

Still not the tool for exploration, stylistic flair, training LORAs, or precise product shots.

01 Midjourney Exploratory work. Stylistic flair.
02 Reve Dynamic, fashion edge with style.
03 Flux Open weights. Fine-tune flexibility.
04 Nano Banana Pro Product shots. Precise, clean outputs.
Aerial lawn bowls — GPT Image 2.
08 · Price

Cheap enough
to experiment.

Go low early. Go high late.

Quality

Size

$0.01

Per image · 1024×1024 · Low

100 images

$1.00

1,000 images

$10.00

Man in blue room — GPT Image 2.
09 · Prompt rewriter

A system prompt
that does the work.

Add this prompt optimiser to a folder in H+Co's Chat UI. Drop in a raw idea, get back a production-ready gpt-image-2 prompt with size, quality and exclusions baked in.

Download system prompt ↓ Open chatgpt.hco.dev →
10 · The catch

Its strength
is its weakness.

Lazy briefs come back lazy. The tool doesn't hide shit thinking. And this is great news for us because we all think. When you can describe a scene, a subject, a detail, a constraint and whatever else, these tools are getting so much closer to meeting us where we are. tbc.

Try it on fal.ai →
Prompt copied