Create with Necmi

Change clothes, keep the person#

Reply to the person’s photo with /image, select FLUX, then send:

Change only the outfit to a navy linen suit and a white shirt. Keep the same face, hairstyle, body proportions, pose, camera angle, background and lighting. Natural fabric folds, fully clothed.

Use a photo you have permission to edit. Check that the face and hands have not changed.

Create a product photograph#

Send /image, choose a model and format, then send:

A studio photograph of a handmade silver ring on pale stone. Soft daylight from the left, realistic metal reflections, macro detail, clean background, no added text or logos.

To preserve a real product, reply to its photo instead of generating from text alone.

Animate a still image#

Reply to an image with /video and choose an available model:

A 5-second clip. The dog slowly turns its head toward the camera and blinks. Keep the same coat markings and harness. Locked camera, subtle movement, no scene change.

Small, specific movements are easier to inspect than several actions in one short clip.

Replace faces or a whole character in a video#

In Playground, choose Replace people in video to open the person-selection flow alongside your chat. You can also open Replace in video directly. Choose the mode before preparing your clip. Available modes depend on your bot and account permissions.

1. Upload or link a video

   Upload MP4, MOV or WebM up to 20 MB, or paste a supported public YouTube, Instagram, TikTok or X link. Choose Load video. For a public link, wait for the full source to become playable before choosing a trim. The source must fit the import limits. Loading does not find faces or generate a video; returning to this page link keeps you at the trim step until you run face analysis. Private or login-protected links cannot be imported.

2. Trim the video

   Play or seek through the source, then use Start here and End here, or enter the times directly. Use Preview selected range to check the selection before finding faces. Times refer to the original source. Leave the end blank to use the rest of the actual video track, even if a public link reports a rounded duration. The selected range determines the output length, within one video frame. Face mode allows the configured limit, up to 60 seconds. Whole-character mode requires 0.5–5 seconds, with the intended person visible in that range’s first frame. If the range exceeds the mode’s limit, choose a shorter end time; the app does not silently shorten it.

3. Select the people to change

   Find faces, then choose up to five in face mode. If someone is missing, pause on a clear face and scan that frame. Whole-character mode scans only the opening frame and allows one person. If the selection is wrong or missing, go back, change the trim and scan again. Scanning again replaces the previous suggestions and selections.

4. Choose a character or add a photo

   Assign a private saved character to each selected person, or use Add a new character to save a name and a JPEG, PNG or WebP photo up to 5 MB. Whole-character mode uses one reference and asks for its type: Person, Dog, Cat or Other. A clear face and visible body help the model follow the reference.

5. Review the instructions and preview

   Face mode shows the intended changes for review. In whole-character mode, edit the prefilled instructions if needed, then choose Prepare preview. Wait for both private images: the replacement identity in the video scene, and the original scene with the selected mask. Check likeness, expression, body, hands or paws, and whether the mask covers only the intended person. Use Back to correct the reference, selection or clip if either image is wrong. You must review both images and check the approval box before generating.

6. Generate and inspect the video

   Confirm permission to use the photos and video, then choose Generate video. The request enters the queue. Keep the page link to return to its status and download the finished video. Inspect the entire result; a correct still preview does not guarantee correct motion throughout the clip.

Changing the prompt, reference, character type, selected person, trim, quality or mode clears preview approval. Prepare and review a new preview before generating. Returning to a saved preview link restores its images and choices, but you must review and approve it again. Prepared clips and previews are private and expire after 24 hours; saved characters remain until you delete them.

To animate the reference photo’s scene in the main composer, choose Copy movement. Add the reference photo, then upload a driving video or paste a supported public link. Set the start and optional end time before sending, using the same blank-end behavior described above. The selected range must fit the bot’s configured Wan limit, up to 60 seconds; longer ranges are rejected, not shortened. The result retains the selected video’s audio. Image thumbnails and playable upload previews help you check the inputs. The editable template describes what to preserve; adapt it to the expression, movement and character you want. Switching to Copy movement selects Wan and preserves your edited scene text.

Ordinary Video mode, including MiniMax H3, does not copy a second video’s movement. If you add a driving video there, sending is blocked: choose Copy movement for the photo’s scene or Replace people in video for the video’s scene.

Auto format keeps the reference photo’s proportions for photo animation. Video replacement keeps the driving clip’s proportions. The model may resize or pad inputs to fit its supported canvas. Auto style preserves the source appearance without adding a cinematic, realistic or illustration preset. Choose a different format or style when you want to override those defaults.

To copy movement while keeping the photo’s scene in Telegram, use /animate --model photo -s 10, then send the photo and driving video. This uses the first ten seconds, or the whole source if shorter. See the trim command reference. For a supported public video link, follow the bot’s import and permission prompt.

Whole-character mode starts with editable instructions for identity, expression, movement and anatomy. Keep them or adapt them to the reference. For example, to replace the selected person with your dog:

Use the reference dog’s round dark eyes, short muzzle, floppy ears, curly white fur and red harness. Follow the selected person’s expression, gaze and movement timing while keeping the dog’s natural compact body, short legs and head-to-body proportions. Keep furry dog paws and consistent coat markings. Preserve the other people, camera movement and background outside the selected mask, and keep the original audio.

The instructions describe the intended result. They cannot guarantee identity, accurate hands or paws, body scale, or faithful expressions and performance. Human-to-animal motion is especially difficult. Check the preview and the finished short video for likeness, anatomy, motion, flicker and mask edges.

Edit a video’s appearance#

Use /editvideo with your clip:

Change the jacket to forest green. Preserve the person’s face, movement, camera path, room and original lighting. Keep the jacket’s color consistent throughout the clip.

A targeted edit is easier to evaluate than changing the person, clothes, setting and motion at once.