Start a project
WorkBlogResourcesAboutContact

Blog/Failures and fixes

Your Vertical Flyer Fails Before Higgsfield Generates an Image

Higgsfield rejected a 4:5 flyer because direct GPT Image 2 generation supports different ratios than product photoshoots.

By Jeff Zambrano··4 min read

Your Vertical Flyer Fails Before Higgsfield Generates an Image

Error: Invalid values: aspect_ratio=4:5

The flyer failed before Higgsfield generated an image.

The complete error pointed to the accepted values:

Error: Invalid values: aspect_ratio=4:5 (allowed: 1:1,4:3,3:4,16:9,9:16,3:2,2:3)
Hint: Run: hf model get gpt_image_2

I had used 4:5 because it is the standard vertical feed shape I expected for a social piece. Higgsfield supports that ratio in one GPT Image 2 route.

It does not support it in the route I was using.

The route controls the allowed ratios

The mistaken assumption was that aspect ratios belonged to the model. Both commands ended in gpt_image_2, so I treated the ratio list as a model-wide capability.

The accepted ratios actually changed with the subcommand.

Route Accepted aspect ratios
generate create gpt_image_2 1:1, 4:3, 3:4, 2:3, 3:2, 9:16, 16:9
product-photoshoot create The same ratios, plus 4:5 and 5:4

I had seen 4:5 in the product photoshoot route and carried it into direct generation. The server rejected it before putting the request in the queue.

This is easy to miss because both routes use the same image model. The product photoshoot documentation can look like a complete list of what that model accepts. It is only a list for that route.

Printed text determined which route I needed

The flyer contained printed text. That meant I needed generate create, where I could provide the full written prompt.

The product photoshoot route did not let me write that prompt. Switching routes only to preserve 4:5 would have removed the control the flyer required.

The working answer was 3:4.

That ratio is accepted by generate create gpt_image_2. It occupies nearly the same amount of vertical feed space and does not require a crop.

For a pure product photograph without printed text, the choice is different. That piece can use:

product-photoshoot --mode lifestyle_scene

In that route, 4:5 is valid.

The content of the piece therefore decides the route, and the route decides the ratio. Starting with the desired ratio and forcing every piece through it reverses that relationship.

The failure did not spend credits

The server validated the aspect ratio before queueing the generation. No image was generated, so the failed request cost zero credits.

The time cost can still grow quickly.

If ten vertical pieces are launched in parallel with the same invalid ratio, all ten can die on the same validation error. There is no creative output to review and no variation in the failures. The batch only repeats the same unsupported parameter.

That is why this check belongs before parallel generation.

The error itself gives a direct hint:

hf model get gpt_image_2

Before launching a batch, I use JSON:

higgsfield model get gpt_image_2 --json

I inspect the accepted values for the specific command path before launching the set. I do not rely on the ratio I used in another Higgsfield workflow, even when the model name matches.

auto is not a safe escape hatch

This route-specific behavior also matches another aspect-ratio failure in Higgsfield: --aspect_ratio auto can fail in nano_banana_2.

The lesson is broader than choosing between 4:5 and 3:4. Aspect-ratio enums are specific to the path that receives them. They are not inherited reliably across commands just because the commands share a model or belong to the same platform.

For a flyer, I now decide in this order.

Does the image require printed text? If yes, it goes through direct generation with a ratio supported there. For a vertical feed piece, that ratio is 3:4.

Is it a pure product photograph without printed text? If yes, the product photoshoot route can use 4:5.

Before a batch, I query the model information and confirm the ratio in the exact route I intend to call.

The rule I use now

I validate aspect ratio by subcommand, not by model.

4:5 working in product-photoshoot create says nothing about whether generate create gpt_image_2 will accept it. For a vertical flyer with printed text, I use 3:4. For a product image without text, I can use the product photoshoot route and keep 4:5.

Before you launch your next vertical batch, run the model inspection command and compare the ratio against the direct generation list. That check costs nothing and can prevent every item in the batch from failing before generation.


Every fix in this article was paid for with a mistake first. Better on my film than on your brand.

ShareXLinkedInWhatsApp
Jeff Zambrano

Jeff Zambrano is a Madrid-based AI film director. His debut, Borrowed Wounds, reached the Top 50 of the Higgsfield Global Film Festival out of 6,700 films. He writes these notes from his own productions.

WhatsApp