Ideogram 4: The 4.5 Editing Update for Creators

Searching for Ideogram 4 can lead to three different decisions: which image model to start with, where that model is available, and whether a later edit will preserve an approved creator or product. Those are not interchangeable questions. A new generation can look compelling while failing a campaign's identity contract. This guide separates the documented access paths and provides a protected-region worksheet for a controlled edit. It does not present a vendor demonstration as our own test.
Create an authorized creator still in APOB before evaluating a targeted edit.
The October 2, 2026 Runway changelog is an access checkpoint, not a launch date for Ideogram 4.5 and not proof that every Ideogram editing control is available through Runway. If the job is to revise one feature of a saved campaign image, verify the editing route separately. Keep the original, the instruction, and the reviewed export together so a later reviewer can tell what changed.
Read the Ideogram 4 family map before choosing a workflow
First record model, interface, task, input, edit controls, account access, dimensions, and check date. A model name is not a capability guarantee.
What 4.0 and 4.5 name
Ideogram 4 is a model family, not a universal interface. The Ideogram 4.5 model page shows generation and targeted-edit examples, not proof that your face, garment lettering, or product label will remain unchanged. Record the model label and controls visible in the actual interface. An integration with fewer controls is a different production route.
Generation versus Precise Edit
Generation asks for a new image. A precise edit asks for a bounded change to an existing one. The Ideogram API overview documents a Precise Edit route with masks and reference-image inputs; it should not be inferred from a third-party image-generation listing. For an approved influencer portrait, the difference is consequential: a new image can change identity cues, whereas an edit request should specify both the changed area and the protected areas. Neither label eliminates the need to inspect the exported pixels.
Where the current documentation applies
Put a source URL beside each claimed control. The Runway changelog speaks to its MCP integration; Ideogram's model page describes Ideogram's product presentation; its API documentation describes API operations. They are different contexts. Check an account's actual UI and plan before promising a client that a given route can mask a jacket, preserve face geometry, or revise text. Document unsupported steps as unknown rather than inventing a workaround that changes the approved asset.
Use the October MCP update as an access checkpoint
Preflight the actual account: can it complete this authorized revision at the required size? Record its model picker, input fields, and export settings.
The October 2 Runway entry
Runway's October 2 entry says Ideogram 4.5 image generation was added to Runway MCP. That is a dated distribution update. It does not date Ideogram 4.5's original release, does not say APOB embeds Ideogram, and does not document Precise Edit through MCP. A comparison article that collapses these into “Ideogram 4.5 editing launched on October 2 in APOB” would be wrong on both timing and product boundary.
Paid-plan and client checks
The same Runway entry associates MCP access with paid plans. Confirm current eligibility in the actual account and client before building a handoff around it. Check whether the surface accepts an existing image, a mask, and reference images or only a prompt for image generation. If a feature is absent, record “not available in this tested route”; do not treat the absence as proof the vendor lacks it everywhere. Also note whether the file can be exported at the dimensions and format your campaign needs.
Why editing needs its own access check
A text-to-image endpoint may produce a striking alternative still yet fail the assignment to change only a product's backdrop. A real editing test needs the saved source image, allowed edit area, protected-region list, exact instruction, output file, and a reviewer. If the approved route cannot preserve these artifacts, choose a different editor or hold the revision. APOB's AI Photo Editor may be relevant to an upstream or separate image-editing workflow, but it should not be described as an Ideogram gateway without documented evidence.
Audit local edits with an unchanged-region worksheet
Compare whether the requested change happened without damaging approved regions. Use the same source and acceptance question. This is a blank worksheet, not a measured result.
Field | Record before running | Record after export |
|---|---|---|
Source identity | File ID, owner, consent scope, dimensions | Original retained unchanged |
Requested change | One visible target and permitted area | Pass, partial, or fail with crop |
Protected zones | Face, hair, garment, logo, product geometry, copy | Any defect and frame/coordinate |
Inputs | Prompt, mask, reference file IDs, model and route | Exact settings or unavailable control |
Export | Required aspect ratio, format, color profile | Actual dimensions and file checksum |
Decision | Reviewer and acceptance threshold | Accept, repair, or retain source |
Requested change and protected zones
Limit the request to one scene variable—for example, replace only the neutral studio wall with a warm interior while preserving the saved fictional host, hair silhouette, shirt seam, and product bottle. Draw the editable area; list protected areas in words a reviewer can observe. “Keep it consistent” is too vague. A protected zone can fail even when the output looks attractive: a shifted logo, extra finger, or subtly different eye shape is still a campaign defect.
Mask and reference provenance
If a mask is available, save it with the source image and note whether white, black, or transparency indicates the editable area in that exact interface. Record all reference files and their rights. Ideogram's API documentation describes reference-image support for Precise Edit, but a particular integration might expose a smaller subset. Never imply that a reference uploaded to one provider is automatically cleared for use on another. If real-person likeness or branded material is involved, ensure the permission covers the new processing and distribution route.
Boundary and pixel-difference log
Inspect the boundary at 100% and at intended feed size. Note halos, mismatched shadows, fabric discontinuities, warped product edges, and unwanted text changes. A pixel-difference image can help locate changes but cannot by itself judge whether a face remains recognizable or whether a legal label is truthful. Put the original and output on the same canvas, annotate the defect with coordinates, and retain both. Do not print a “0% drift” score unless the images and computation are attached and the metric's blind spots are explained.
Carry an APOB creator still through one bounded revision
The APOB AI Influencer Generator and AI Image Generator can supply the authorized creator source. Preserve it before testing a separate editing route; neither page promises a native Ideogram integration.
Authorized saved persona source
Choose a fictional host or a person whose consent covers AI imagery and the intended channels. Save the unmodified source, a rights note, and a short identity card: hairline, eye shape, skin-tone reference, defining accessories, approved wardrobe, and the product's real features. If the input was itself generated, preserve the prompt and model/version where available. An attractive exported still is not enough if a reviewer cannot trace it back to the approved source.
One scene variable per instruction
Example instruction: “Change the background to a daylight kitchen; keep the host's face, blue jacket, bottle geometry, and all printed text unchanged.” This is a proposed test instruction, not a reported result. Run one version at a time. If a second revision is needed, start from the approved master or document why an intermediate output became the new source. Otherwise errors can compound and the team may lose track of which face or bottle was actually signed off.
External output and revision naming
Use filenames such as hostA_source_v01.png, hostA_kitchen-edit_v01.png, and hostA_kitchen-edit_v02.png. Record the route and date in the manifest, not only in memory. Keep the original as read-only; place new outputs in a separate review folder. If the editor strips metadata, add the source and export details to the worksheet. This structure makes a later “change only the wall color” request possible without re-creating a face from scratch.
Decide which result can enter the campaign master
Review actual exports against the brief at placement size. A changed face or invented product claim fails even when the image looks polished. If none passes, retain the source.
Identity and product truth
Compare the face, proportions, garment, product color, package count, and material details to the master. Inspect protected areas at both 100% and feed size. A new reflection can suggest a different product shape; a hand added to the scene can change the implied use. Treat these as factual or identity issues, not cosmetic preferences. Record the exact crop that shows the problem so the next pass has a bounded correction target.
Text and permission checks
Read every visible label, price, subtitle, and brand name. Generated text is not an approved specification, even when it looks plausible. Confirm the rights for the source portrait, references, product art, and destination channels; mark synthetic portrayal if required by the placement. If an integration's terms or export rights are unclear, ask the account owner for the current agreement rather than turning a model availability note into a legal conclusion.
Accept, repair or retain the source
Accept only when the requested change is visible, protected zones and factual details pass, dimensions match the placement, and rights are documented. Repair when a localized defect can be corrected without reopening identity or product uncertainty; name the specific region and a new review step. Retain the source when the edited output drifts, access is unavailable, or a clean handoff cannot be proven. The worksheet is the reusable output of this guide. It supports a real test; it is not a substitute for one.
Start with an authorized APOB campaign still, then record the independent editing route and compare its actual export with the protected-region worksheet before replacing a campaign master.
Related workflow: Leonardo AI Alternative for Fashion Lookbooks















