Generation and Account Details

Detailed, verified operating rules for GLMImage generation and account flows.

Generation and Account Details

Authentication and draft recovery

Generation requires a signed-in account. When a guest selects Generate, GLMImage saves the current scene, prompt, workflow, variant, quantity, output size, custom dimensions, and uploaded reference URLs in session storage. After successful authentication, the draft remains available. GLMImage does not automatically spend credits or start the task after sign-in; the user confirms Generate again.

Trial and credit rules

  • Every new account receives 20 credits, enough for 2 trial images.
  • Image generation currently costs 10 credits per image.
  • Text-to-image can submit multiple independently billed tasks when the account and balance allow it.
  • Image-to-image submits one output task at a time.
  • When a trial is exhausted or credits are insufficient, the upgrade flow opens with a categorical reason.

Output validation

Output dimensions must be 256–1536 pixels on each side and divisible by 32. The browser checks the value before submission, and the API validates it again before contacting a provider. Invalid values do not create a task.

For image-to-image workflows, the required reference count comes from the selected workflow. The current global upper bound is four images. References must finish uploading successfully before task creation.

Models and providers

GLMImage is a product brand, not a promise that every result comes from a single model. The server resolves the production model from the trusted routing configuration; a model value supplied by the browser cannot override it. The resolved provider and model are stored with the task for later verification.

Failure handling

Task-creation failures use stable categories such as:

  • AUTH_REQUIRED
  • TRIAL_EXHAUSTED
  • INSUFFICIENT_CREDITS
  • INVALID_REFERENCE_COUNT
  • INVALID_REFERENCE_IMAGE
  • INVALID_OUTPUT_SETTINGS
  • GENERATION_FAILED

Raw provider responses, prompts, image URLs, file names, email addresses, and stack traces are not sent as PostHog event properties. Public error messages are intentionally generic; detailed provider errors remain in controlled server logs.

Only a generated result can be submitted to the public Gallery, and only after the signed-in user explicitly selects Publish. Reference images remain private. Workflow galleries show approved database records and do not insert fabricated ratings, testimonials, or placeholder works.

Troubleshooting checklist

  1. Confirm the account is signed in.
  2. Confirm the prompt is present when the workflow requires it.
  3. Check that custom dimensions are in range and divisible by 32.
  4. Wait for every required reference image to finish uploading.
  5. Confirm the reference count is within the workflow limit.
  6. Check the available credit balance or trial eligibility.
  7. Retry once if the service reports a network or general task-creation error.
  8. If the same categorical error persists, contact support with the task ID; do not send private prompts or source images unless support explicitly needs them.

On this page