Choose the kind of change before choosing settings
Higgsfield distinguishes Motion Transfer, which rebuilds a scene around the source movement, from Object Swap, which targets selected elements. For a duo remake, decide whether you want the whole performance recast or a more limited replacement. The decision affects how you describe the result and what you expect to remain from the original.
Write the goal in plain language before opening controls. For example, you might want two authorized people to occupy the source performers' positions while preserving the camera and timing. If you also want a new location, different outfits and a different style, record those as additional changes rather than assuming they are part of a simple face replacement.
Use the active generator as the final reference for supported inputs. A tutorial showing a similarly named video model may be describing a different workflow. Starting from the official Genjutsu page reduces the chance that you upload a source into a mode designed only for still images or text.
Prepare a source clip you can actually use
Choose a continuous performance excerpt with both subjects visible. Inspect the first and last frames as well as the action in between. If the excerpt starts during an abrupt cut or ends halfway through a gesture, the transformed output inherits a difficult editorial boundary before any identity work begins.
Keep the intended frame in mind. A wide source that places performers near opposite edges needs deliberate treatment before becoming a narrow post. Do not crop away half of the scene merely to match a vertical target. Decide whether you will preserve the full composition or use a source already framed for the destination.
Use footage you are authorized to process. A link to a public performance identifies a reference but does not supply an upload-ready file or establish permission for every use. This guide does not offer a download bypass. If the original file is unavailable, choose an authorized alternative or a workflow that supplies suitable source material.
Keep an untouched copy of the excerpt and note its duration. That is your comparison reference when you evaluate the result. It also prevents later trimming, soundtrack replacement or repeated re-encoding from obscuring whether a problem originated in the source or during generation.
Make the two portrait roles easy to distinguish
Prepare one clear image for each intended performer. Give the files simple working names that identify the left and right roles, then confirm the uploaded thumbnails match those names. Similar filenames from a phone library are easy to confuse, especially when you have several edited copies of the same photograph.
Choose references that provide the information your request needs. A headshot can identify a face, but it does not fully specify an outfit or body silhouette. If your planned change includes clothing or a broader character design, inspect the current reference controls and provide suitable material rather than assuming the face image contains those details.
More references are not automatically better. Conflicting ages, hairstyles or clothing can make the desired identity less clear. Start with the smallest coherent set that describes the project. Keep any additional image's purpose explicit so you can remove it if the output starts combining incompatible features.
The documented Hotel Lobby Higgsfield sequence
Higgsfield's official walkthrough describes choosing a mode, adding a source video, adding visual references when needed, then selecting a preset or supplying a short description. Generate after reviewing the settings and compare the output with the source. A full written prompt is not required when a supported preset already expresses the change.
The live documentation currently describes source videos from 4 to 30 seconds and up to 30 reference images. Older search summaries exposed different limits during this review. Check the actual upload form before preparing a large batch; the interface's accepted duration and reference count are the practical constraints for your run.
For a Hotel Lobby Higgsfield experiment, a concise description should associate each portrait with its intended performer and identify the motion or framing to retain. Avoid asking for an entirely new choreography while also demanding exact preservation of the source. The prompt guide explains this reference-role approach in more detail.
Check the quote for this run, not a price from a short tutorial
Genjutsu displays a credit cost before generation. Review it after selecting the duration and output settings you actually intend to use. The official walkthrough contains historical examples tied to length and resolution, but a published example is not a promise that your current account will show the same charge or cash conversion.
Separate a generation charge from the cash needed to obtain credits. Existing subscribers may already have an allowance; a new user may need a plan or pack. Also separate subscription credits from other balances because their expiration rules can differ. Use the current billing screen for the purchase decision.
If your budget covers only one attempt, finish the input review before submitting. Check that both portraits loaded, the correct source excerpt is selected, and the chosen mode matches your intended change. Do not plan around an assumption that an aesthetically disappointing result will automatically be refunded.
Compare the output to the source at matching moments
Watch the source and result at the opening, a strong gesture, any overlap between people, and the closing moment. Those checkpoints reveal whether the transformed scene preserves the action you cared about. An appealing first frame can hide later identity drift, changes in body proportions or a sudden rearrangement of the scene.
Check the two people separately. Follow the left person through the whole clip, then repeat for the right. Looking at both at once can make a brief face substitution easy to miss because your attention naturally follows the more active performer. Note the exact moment of a defect before deciding what to change.
Listen to the exported file. This guide does not assume that every Genjutsu mode preserves the source soundtrack in the same way. Confirm whether the result contains the expected audio, whether it starts at the correct point and whether the visible performance fits it. If needed, finish sound in an editor using material appropriate for your intended use.
Use a specific correction instead of restarting the whole project
If the wrong person occupies a role, check reference association first. If identities are right but motion is awkward, examine the source excerpt and any conflicting instruction. If the crop is wrong, revise the composition. These are separate problems, and changing every input at once makes the next result difficult to interpret.
Name your versions by the change they test, such as clearer-right-portrait or wider-source-crop. Keep the note beside the files, not only in your memory. After several versions, similar thumbnails make it difficult to remember which setup produced a particular problem. A useful comparison depends on being able to connect an output to the inputs that created it.
Before a new attempt, write one acceptance condition. For example, the right performer must keep the same face through the turn, or both people must remain visible at the widest gesture. Review that condition first when the next file arrives. If it passes but another defect appears, record both rather than declaring the whole change successful from its opening frame.
Keep the accepted output and source together when you move to editing. Add captions or make a short trim there rather than requesting another full transformation just to change presentation. Preserve an unedited export so you can return to it if a later effect or soundtrack adjustment introduces a defect.
Choose this route when control over your own source footage is part of the project. For the fixed 15-second human-duo workflow on this site, the main creation guide explains the narrower set of decisions. Neither route removes the need to review identity, timing and sound before sharing.