New on the blog: why AI belongs in the feed →
View Categories

Image Feed Settings

Image feeds are created under Form Settings -> AI Image and use OpenAI.

Model lifecycle #

GPT Image 2.6 Flare is the recommended default for a new feed. GPT Image 2.6 Sunburst is the model of the pair for detailed visual work and is worth choosing when precision matters more than speed; OpenAI charges both at the same token rates as GPT Image 2.5. Both 2.6 models accept Extra High and Maximum quality. Custom sizes remain limited to the model IDs whose support OpenAI explicitly documents; choose a GPT Image 2.5 model when a feed needs a custom size.

GPT Image 2.5 Flare and GPT Image 2.5 Sunburst remain current and selectable. OpenAI has not replaced them with 2.6, and the historical GPT Image 1 shutdown row still names them as its replacement. A feed that uses 2.5 does not show a migration notice.

GPT Image 2 is deprecated and retires on February 10, 2027. It is not offered for new selections. A feed that already uses it keeps showing it with a warning until that date, then stops generating images until an administrator explicitly selects and saves a supported model. OpenAI names GPT Image 2.6 Flare and GPT Image 2.6 Sunburst as the replacements. The plugin never moves a feed to a different model on its own.

GPT Image 1 (gpt-image-1) retired on October 23, 2026. An existing custom feed using that ID cannot generate images until an administrator selects and saves a supported model. GPT Image 2.6 Flare is the recommended migration option.

GPT Image 1.5 and GPT Image 1 Mini are deprecated and are not offered for new selections. A feed that already uses one keeps showing it with a warning until OpenAI’s December 1, 2026 retirement date. On that date the feed stops generating images until an administrator explicitly selects and saves a supported model. The plugin never moves a feed to a different model on its own, because model behavior, capabilities, and pricing can differ.

Mode #

Requires Premium or higher. Generate a new image creates an image from the prompt and is the default. Edit an uploaded image changes an image that was uploaded with the entry.

Editing has these rules:

  • It runs after the form is submitted, as a Gravity Flow step, or as an administrator rerun. It is not available from a frontend button, because the image to edit is saved with the entry.
  • The edited image is saved to the Target File Upload Field, which must be a different field from the source image and the markup overlay. The uploaded originals are never changed.
  • SVG output is not available. PNG, JPEG, WebP and, on GPT Image 2.6, Animated GIF are.
  • GIF frame count and duration are not sent with an edit.

Image To Edit #

The File Upload field that holds the source image. It can be a file the visitor uploaded or an image an earlier AI image feed saved to the entry, so one feed can generate an image and a later feed can edit it. The file must be PNG, JPEG or WebP and no larger than 50 MB. For a multi-file field, the first file is used. If the entry has no file in that field, the feed fails with an entry note and no request is sent.

The plugin reads only files inside this form’s Gravity Forms upload folder. It does not fetch images by URL.

Markup Overlay (optional) #

A File Upload field that holds a drawing over the image, such as a circle, an arrow or a scribble, showing where to make the change. Requires a GPT Image 2.5 or 2.6 model.

The overlay must be:

  • a PNG with a transparent background. Draw the marks on a transparent layer; do not draw on a copy of the photo.
  • exactly the same width and height as the image to edit, after any rotation stored in a phone photo is applied.
  • at most 16,777,216 pixels and 50 MB. The image to edit has the same pixel limit when an overlay is used.

These are checked before the request is sent, and the entry note says which rule failed. When an entry has no overlay file, the image is edited from the prompt alone.

In the prompt, refer to the marks and say what should happen to them, for example: “Recolor the vase inside the red circle to cobalt blue. Keep everything else unchanged. Remove the red circle from the finished image.”

There is no drawing tool in the form. The visitor or an administrator uploads an overlay made in an image editor.

Prompt #

Describe the image to generate. The merge-tag button inserts current form values. Good prompts state the subject, setting, composition, visual style, colors, orientation, intended use, text that must appear, and important exclusions.

Example: Create a clean square event illustration featuring {Event subject:3}. Use {Brand colors:4}, leave clear space at the top for a headline, and do not include logos or written text.

All merged values are sent to OpenAI. Provider safety policies can reject a prompt even when its format is valid.

Target File Upload Field #

Choose a File Upload field from the current form. The generated image is saved there so Gravity Forms and connected add-ons can treat it as an entry file.

For an after-submission feed, the target File Upload field must be optional. Gravity Forms checks required fields before the feed can generate the missing file. Frontend generation also needs the target field so the verified temporary image can be associated with the submitted entry.

Conditionally Hidden Target #

Shown for frontend image feeds and enabled by default. Leave Attach the generated image even when conditional logic hides the target field selected when the generated image is an output of the feed rather than a visitor-supplied upload. Gravity Forms normally omits conditionally hidden fields while creating an entry; the plugin attaches the verified generated image immediately after the entry is saved so later feeds and notifications receive it.

Clear this setting when hiding the File Upload field should also prevent the generated image from being attached. The visitor can still generate and preview the image, but it is not added to the entry if the target is hidden when the form is submitted. This setting does not change the field’s conditional logic or make it visible.

Gravity Forms must create an entry before the plugin can attach the image. If the hidden target is the only field with a value, Gravity Forms treats the submission as empty and displays At least one field must be filled out. Ensure the form submits a value from at least one other visible field; the generated image cannot satisfy this check because its hidden target is attached only after the entry is created.

Size #

  • Automatic sends auto; OpenAI chooses the size according to the selected model and service behavior.
  • 1024 x 1024 creates a square request and is the default for a new feed.
  • 1536 x 1024 creates a landscape request.
  • 1024 x 1536 creates a portrait request.
  • Custom size reveals Custom Size. It requires GPT Image 2.5 Flare, GPT Image 2.5 Sunburst, or an existing GPT Image 2 selection. New feeds cannot select GPT Image 2.

Automatic does not mean the plugin inspects the form or preview area. The decision is delegated to OpenAI.

Custom Size #

Enter WIDTHxHEIGHT, such as 1280x720. Both edges must be divisible by 16. The aspect ratio cannot exceed 3:1, no edge can exceed 3,840 pixels, and the total must be between 655,360 and 8,294,400 pixels. This setting is accepted only for GPT Image 2.5 Flare, GPT Image 2.5 Sunburst, and GPT Image 2. OpenAI treats resolutions above 2560 x 1440 as experimental.

Larger images can take longer and may cost more. Confirm current provider pricing before using large dimensions on public forms.

Quality #

  • Automatic sends auto; OpenAI chooses the quality.
  • Low, Medium, and High request an explicit quality level and are accepted by every image model.
  • Extra High and Maximum request the two higher levels supported by GPT Image 2.5 and GPT Image 2.6 Flare and Sunburst. Choosing one with any other model reports an error on the Quality setting when the feed is saved.

OpenAI’s image generation guide lists both families’ supported quality levels in its cost calculator.

Higher quality can increase generation time and cost. OpenAI bills image output per token, and Extra High and Maximum spend several times what High does on the same prompt, so confirm current pricing before putting either on a public form. Automatic does not inspect the prompt or site design locally; it delegates the choice to the provider.

Output Format #

Choose PNG, JPEG, WebP, Animated GIF, or SVG vector. PNG is the default and is useful for lossless graphics. JPEG is often smaller for photographic images. WebP can offer smaller modern web images but may be less suitable for external systems that expect PNG or JPEG.

Animated GIF requires GPT Image 2.6 Flare or Sunburst and provider access to the GIF preview. Describe the subject, motion, and desired loop in the prompt. Choose Automatic size when starting an animation; OpenAI applies model-specific frame and total pixel limits, and a fixed size may require fewer frames. The plugin preserves the returned GIF bytes through previews and entry attachments, so it does not flatten the animation into a still image. The target File Upload field must allow the gif extension.

SVG vector #

SVG output requires GPT Image 2.6 Flare or Sunburst and SVG access on your OpenAI project. It suits logos, icons and flat illustrations, and is not available in Edit mode or with previews while generating. The provider first draws a reference image and then a second model writes the vector file. The ledger’s token totals include both; Quality and Complexity apply to the reference image.

SVG Author Model and SVG Author Effort appear when the output format is SVG. Leave the model blank unless SVG model selection is enabled on your OpenAI project. With it enabled, OpenAI requires a model and rejects the request without one; without it, OpenAI rejects a request that names one. Enter the model ID exactly as OpenAI lists it for your project; the plugin has no list of valid IDs to check it against. Effort sets how much reasoning that model uses and is sent only together with a model.

An SVG file is a document a browser can run, so the plugin checks every generated SVG before saving it and stores only its own re-written copy. The file is rejected, and nothing is saved, if it contains anything other than static drawing content: scripts, event handlers, links, embedded or external images, animation, foreign content, references to anything outside the file, a document type or entities, or XML that is not well formed. A rejection appears as the feed’s error with the reason; generate the image again.

WordPress does not allow SVG uploads, and the plugin does not change that. It vouches only for its own generated and checked files during the form submission that attaches them. Previews load the file by URL as an image and never insert its markup into the page. If the target File Upload field limits file extensions, add svg to its list.

Complexity #

GPT Image 2.6 Flare and Sunburst accept Low, Medium, or High planning complexity. This controls planning before rendering separately from Quality. Higher complexity can help with dense text and layouts, but can increase latency or change more details.

Provider default leaves the parameter out and currently uses High. Selecting High explicitly sends High even if the provider later changes its default. Feeds saved before this control was added keep that behavior. A saved complexity value is ignored for models outside the reviewed support list, including older models and unknown custom IDs.

GIF Frame Count And Duration #

These controls appear only when Output Format is Animated GIF. Frame Count accepts 2 through 16; leaving it empty uses the provider default of 9. Frame Duration accepts 20 through 2,000 milliseconds per frame and rounds to the nearest 10 milliseconds; leaving it empty uses 200. Changing timing changes playback speed. More frames can increase generation time and cost.

The controls are validated before generation and sent only for GIF output. Switching to a still format leaves saved frame settings inactive. Provider-specific combined size limits are enforced by OpenAI; use Automatic size or reduce the frame count if the provider rejects the combination.

See OpenAI’s GIF guide for the documented frame controls and preview requirements in the API reference.

Content Filter #

OpenAI screens every image prompt against its own content policy, before and after generation. This setting chooses how strictly it does so.

  • Standard (provider default) sends nothing extra and accepts OpenAI’s normal filtering.
  • Less restrictive asks OpenAI to relax filtering to the extent it permits.

Less restrictive is not a bypass. OpenAI’s policies still apply, it can still decline a prompt, and you remain responsible for what your forms ask it to produce. The setting is sent only for models that support it; a custom model ID does not receive it.

Reach for it when legitimate prompts are being declined — descriptions of humanitarian, medical, or emergency work are the common case. Rewriting the prompt usually helps more than changing this setting. See the image generation troubleshooting page.

Content Refusal Message #

The opening sentence shown to the visitor when OpenAI declines a prompt under its content policy. Leave it empty to use the default.

Write it for the route forward this form actually offers. A form whose only option is to try again should not tell visitors to upload a file, and a form that accepts uploads should say so.

A sentence about the likely cause is added after it automatically. Visitors are only ever asked to change their own answers; when the cause came from this feed’s prompt they are told to contact the administrator instead, and the cause is recorded on the Overview tab for you.

Visitors can see this text. Do not put internal notes, provider details, or contact information that should stay private into it.

Preview While Generating #

Shown only for a frontend trigger with PNG, JPEG or WebP output. Choose one to three rough versions to show in the preview area while the image is being drawn. The finished image then replaces them. Only the finished image is stored and attached to the form; a rough version is never saved.

Each rough version adds a small number of image output tokens to the request. If the provider does not accept previews for the selected model or project, the image is generated without them. The setting is Off by default.

Who Can Use The Frontend Button? #

Shown only for a frontend trigger. Public access allows anyone who can view the form to request an image, subject to signed tokens and rate limits. Administrator-only access requires a logged-in user with Gravity Forms full access. See the trigger/access reference.

Stuck? Ask in GravityOps Help #

Search past answers or post your question, tagged "ai-feed".