Source: raw/5_Ways_to_Make_a_Talking_Photo_With_AI_Bring_Any_Picture_to_Life.md — “5 Ways to Make a Talking Photo With AI: Bring Any Picture to Life”, YouTube (youtube.com/watch?v=srOvS-iK-gk), uploaded 2026-09-17, transcript fetched 2026-10-30. Creator: HeyGen official channel (@heygen_official); presenter not named. Vendor content: it points to HeyGen’s free plan twice. Platform: YouTube. Duration: 9:23. Skill level: beginner.

A HeyGen tutorial that starts from one still photo, turns it into a photo avatar, and then shows five different outputs from it: a typed-script talking presenter, the photo performing your own audio recording, a multi-scene video in AI Studio, a short cinematic clip, and a live avatar that answers questions out loud. The useful part is the decision rule at the end: each method fits a different job. All claims come from HeyGen’s own channel and the video shows no quality comparison or cost per method.

Key Takeaways

  • The photo matters more than any later setting. Eyes and mouth clearly visible, face big enough to see, facing roughly toward the camera; a headshot or chest-up photo is “the safest option”. Blurry photos, side profiles and tiny faces give bad results.
  • It doesn’t have to be a real person. Cartoons, animals and invented characters work if the face has clear eyes, a visible mouth and “roughly human proportions”.
  • Permission rule: only upload a photo you own or have permission to use, and the same for uploaded audio. “Don’t try and animate a photo of somebody who didn’t agree to it.”
  • Making the photo avatar: left menu Avatars → New avatar → New character → Upload photo, name it, submit. Design with AI generates a character if you have no photo.
  • The five methods and where each starts:
#MethodWhereBest for (per the source)
1Talking presenter from a typed scriptHover the avatar → Quick create (opens Single scene, one continuous shot)Quick message, social post, introduction, short explanation. “That’s where everybody should start.”
2Photo performs your recordingSame Single scene, but upload an audio file instead of typingLines where delivery matters: a personal message, an emotional line, a character voice
3Full video built around the photoApps → AI Studio → New video; build scene by sceneTutorials, training, presentations, “anything longer than just a short message”
4Cinematic sceneApps → Cinematic shots, photo as referenceShort striking clips for the first few seconds of a video
5Live avatar that answers in real timeSeparate site, liveavatar.comSupport, coaching, website facts: “anywhere where the pre-recorded video wouldn’t work because you wouldn’t know the question in advance”
  • Method 1 controls: type the script, pick a library voice or your own voice clone, optionally Set motion (describe how the avatar moves) and Set expression, choose landscape (YouTube, website) or portrait (TikTok, Reels, Shorts), then generate.
  • Method 2 trade-off: the uploaded recording keeps your timing, tone, emphasis and pauses. “The one typed was faster… but the recorded one feels more like me.”
  • Method 3 structure in the demo: scene 1 introduces, scene 2 cuts away to footage and text, scene 3 brings the presenter back to close. Each scene can change script, voice, captions, background, images and layout. Stock images and video are under Media. Advice: use the presenter only where it helps, “Don’t overdo it.”
  • Method 4 prompt checklist: what the person does, how the camera moves, the lighting, and the overall style. Enhance prompt expands a short prompt. Duration runs 4 to 15 seconds; the presenter leaves it on auto.
  • Method 5 has stricter photo rules: JPEG or PNG, at least 1080p, widescreen or vertical, one person centered, chest up, looking straight at the camera, neutral expression with the mouth closed. A smile showing teeth means picking a different photo. Languages and live-avatar presets are selectable; press Chat now to start.
  • Live avatar availability caveat: the presenter demos one of HeyGen’s public live avatars because “live avatar creation isn’t available in my current workspace just yet” and calls the feature “very new”.

Try It

  1. Pick one photo and check it against both rule sets (general and live-avatar), so the same image works for all five methods.
  2. Start with method 1 (the presenter says it takes about 2 minutes): Avatars → New avatar → New character → Upload photo, then Quick create, type two sentences, pick a voice, generate.
  3. Re-run the same line with method 2 using your own phone recording and compare the two side by side.
  4. For a longer piece, move to AI Studio and alternate presenter scenes with cutaway scenes from Media.
  5. Make one 4 to 15 second cinematic shot for the hook, writing the four prompt elements (action, camera move, lighting, style), and click Enhance prompt if the result is flat. See HeyGen Avatar Shots + Seedance 2.0 for a fuller five-element prompt structure. ^[inferred: the source calls the feature “cinematic shots”; whether it is the same as Avatar Shots is not stated]
  6. Only use a live avatar where questions are unpredictable (support, coaching); pre-recorded methods cover everything with a fixed script.

Open Questions

  • Which animation model each method uses is not stated. HeyGen Personal Avatar Playbook says Avatar IV is the model for photo looks and that Avatar V needs a video look, which suggests methods 1 to 3 run on Avatar IV.
  • Cost and plan limits for each method (credits per minute, cinematic shot pricing, live avatar minutes) are not given; the video only says there is a free plan.
  • “Cinematic shots” vs “Avatar Shots.” The relationship between this Apps entry and the Avatar Shots feature covered elsewhere in the wiki is not confirmed.
  • Live avatar creation availability: not available in the presenter’s workspace at recording time (September 2026); no rollout date given. The presenter points to a separate HeyGen video on live avatars that is not in the vault.
  • Caption fixes applied: “Hijen”, “Hey Genen”, “Hen” → HeyGen; “live aavvatar.com” → liveavatar.com (the URL spelling is not shown on screen in the transcript); “chat GBT” → ChatGPT; “Claw” → Claude.