To keep the same model across AI product photos, reuse an approved identity reference and check a set of outputs together. A saved character, a repeated face reference and virtual try-on can all support consistency; they differ in how you provide the person and garment. Choose the method around the poses, outfits and scenes you need, then test it on that shot list.
Four methods compared
Compare methods by the inputs and the changes your shoot needs. These are workflow distinctions, not measured quality rankings. Saving a reference makes it easier to reuse; it does not establish that one method cannot drift or that another must fail after a particular number of frames.
| Method | Input | Useful for | What still needs review |
|---|---|---|---|
| Saved character | An approved set of images saved as a reusable subject | Repeated campaign work without reselecting the identity input | Face, pose, hands and accessories in each result |
| Face or image reference | The approved identity image attached or reused in generation | New poses and scenes with an identity anchor | Identity under the intended angles and lighting |
| Garment try-on | A model image plus a garment image | Dressing the same person in different products | Garment construction, fit representation and face |
| Video reference | Supported image or video references for the selected video model | Motion from an approved identity or frame | Identity through the sequence and fine-detail drift |
Kive offers reusable characters called with @modelname; Leonardo and Krea also offer custom-model workflows. Midjourney’s current Edit documentation describes up to four reference images and replaces the older Omni Reference flow. Flux Kontext uses image-and-instruction editing. Adobe Firefly’s structure reference guides composition, which should not be confused with an identity guarantee. FASHN’s guide demonstrates both Try On with the same model and Face Reference for new images. Botika provides on-model workflows; check its current model-selection controls for your casting needs. Kling’s reference options depend on the selected video model.
Choose a method using the trial below before comparing subscription costs. The Kive pricing guide estimates about 100 image generations for 1,000 credits, with Basic starting at $20/month or $15/month billed yearly. Editing and video consume the same credit pool.
Method 1: train a character once
A reusable character removes the need to select the identity input again for every shot. The walkthrough uses Kive’s saved-character interface; it does not assume that every product called a character model trains new weights in the same way as a LoRA.
Train a character and hold it across a campaign
Gather the references your tool accepts
Choose one to three clear photos of the same person for Kive, following the custom-model input requirements. Include a useful angle for the shot list and avoid conflicting appearances. Other tools accept different counts; more images are not automatically better.
Train the character
In Kive, open Organize → AI models, choose Character, upload one to three clear photos of the same person and name the model. Follow the custom-model guide for the current setup. Generate a sample before proceeding; the saved name alone is not evidence that the likeness works.
Call the model by name in the brief
Reference the saved character with
@modelname, such as “@ada wearing a tailored blazer in an office.” The character-model guide explains this syntax. Keep the same approved identity input across the shot list, then inspect the outputs rather than assuming the name guarantees identical faces.Pair the character with a saved scene
Consistency has two axes, and the model is only one of them. The other is the room. AI studios store lighting, framing and environment as a reusable preset, so a campaign locks the person and the setting independently: the same character in three studios for three placements, or one studio across a whole collection with the character swapped per market.
Change one variable per frame
Move the scene, or the garment, or the angle - not all three at once. A frame that fails only tells you what broke it if a single variable moved. Batch by scene rather than by shot, because switching scenes is where drift creeps back in.
Run a six-frame drift test
Generate six frames across the placements the campaign actually needs, then view them together as a grid rather than one at a time. Nobody notices inconsistency in a single image and everybody notices it in a grid, so the grid is the test.
Retrain when the shot list outruns the references
A reference set showing only the front may be less reliable at profile angles. If the shot list has profile frames, add a profile reference and train again rather than prompting harder.
Do the plan maths before the campaign, not during it
Count the expected generations and finishing work against the current plan allowance. Kive Basic starts at $20/month for 1,000 credits, or $15/month billed yearly, and also offers a 2,000-credit option. The published estimate is about 100 images per 1,000 credits; output settings, edits and video change the total demand.
Photograph or select your reference images the way you would brief a casting: different angles, even lighting, no heavy shadow across the face. The model inherits whatever ambiguity you give it, and a single flattering three-quarter shot teaches it less than three plain ones.
Character models sit on Kive's paid plans - Basic, Pro and Enterprise - while free accounts get demo models to test the flow. Style models, which capture a look rather than a person, are Pro and Enterprise only (kive.ai/docs). Test the likeness again when you change the garment, location or lighting.
For clothing already listed on Shopify, import its existing product images into Kive through Organize → AI models → Import from Shopify. The Shopify integration builds the product reference from up to four product photos; it does not import or create the person’s identity. Use the garment reference alongside your approved character in the full Kive workflow, and inspect seams, print and fit separately. App Store installs can connect a free workspace you administer; character-model features still depend on your plan.
Methods 2 to 4: reference image, try-on, video
Reference generation, try-on and video workflows can also reuse an approved identity. Their suitability depends on what the shot changes: face, pose, garment or motion. Test the exact operation instead of treating training as a universal quality advantage.
Method 2: reference image, no training
A reference-image workflow attaches the approved person to each generation. Midjourney’s current Edit model supports up to four image references and its documentation describes pinning them for reuse across prompts. This replaces its earlier Omni Reference flow. Flux Kontext supports editing from a source image and instruction. Adobe Firefly’s structure reference guides composition; it is not equivalent to a face-identity control. Keep the same approved identity reference and compare outputs under the poses and lighting you need.
Method 3: garment try-on
Try-on combines a person image with a garment image. FASHN’s workflow explicitly reuses the same model while dressing them in different products, and also offers Face Reference when the composition changes. That makes it relevant to both catalog consistency and garment presentation. Botika provides on-model workflows too; evaluate the available casting controls in the version you use. For listing-specific considerations, see on-model photography without a photoshoot.
Method 4: video reference
A motion workflow needs identity checks across time as well as across still images. With Kling or another video tool, use the reference inputs supported by the selected model and inspect a turn, profile and expression change. A good first frame does not establish that the face or garment stays consistent through the clip.
Reading the table
Choose the operation around the change: reuse a model image for a garment swap, use a face reference for a new scene, or save a character for repeated access in a workspace. Compare setup effort, supported inputs and your trial results. A named saved subject is a workflow feature, not proof of superior identity accuracy.
Where consistency still breaks
No tool holds everything. Three failure modes show up across all of them, and knowing them is the difference between a campaign that ships and one that gets re-shot.
Hands, teeth and jewellery drift long after the face has locked, because they are small, high-detail and rarely well represented in a handful of reference photos. Review them per image rather than trusting the model.
Extreme angle changes stretch any reference. A reference set showing only the front may be less reliable at profile angles. If the campaign needs profile shots, include one in the training set.
Garment detail is not identity. A character model reproduces a person, not the clothes they wore in the reference. For the garment to stay exact, it has to be handled as a product in its own right - which is a different tool and a different reference.
Generated likenesses of real people carry consent and rights obligations that no tool resolves for you. If the reference photos are of a real model, check whether the relevant permissions cover the intended AI use. Check before the campaign, not after.
Why prompt wording cannot hold a face
A text description is a weak identity anchor: changing a scene or pose can change the person that is generated. Reusing an approved image or saved character gives the model a more specific reference. Both still need an output check, especially when the shot changes the face angle, lighting or visible accessories.
A campaign workflow that scales
Sequence the work so the expensive decisions happen once. Train the character first and approve it in isolation, before any campaign context - a face you are not happy with will not improve once a garment and a location are competing for attention.
Then lock the scene. Pick or build the studio for each placement, and generate a single test frame per studio with the character in it. What you are checking at this stage is whether the character survives the lighting, not whether the image is finished.
Only then generate at volume. Batch by studio rather than by shot, and review the small details - hands, jewellery, garment seams - on every frame rather than on a sample. Teams pulling a whole season out of one character will find the campaign-level tooling in our guide to AI lookbook and campaign imagery, and the wider rules that keep colour, type and casting aligned in the AI brand kit guide.
Choose the method that survives your own shot list with acceptable review effort. Keep the references, prompts and outputs together so another teammate can reproduce the setup.
Garment accuracy, prop continuity and hand detail need separate review even when the face looks consistent. A saved character removes some repeated setup; it does not remove the need to inspect each deliverable.
Studios for campaign consistency
References
