The short answer
The difference between a 58th Floor clip that looks like you and one that looks like AI slop traces back to two things: photo quality and photo clarity. Your phone camera is enough. Daylight is enough. A selfie is enough. What matters is that both photos pass the same test: the model can read one face clearly and carry the outfit through 12 seconds of movement without the face drifting or the mouth getting stuck.
Use this checklist before you upload. Each point has a failure mode and a fix.
Checklist: photo requirements
✓ One person per photo, face clearly in view
What passes:
- One person, centered in frame
- Face takes up at least 40% of the photo vertically
- No one else in the background, no arms around shoulders, no group shots
What fails:
- Two people in one photo (the model gets confused about who is who)
- One person cropped at the edge so only half the face is visible
- A group photo you want to crop later (don't—upload single-person photos instead)
How to fix:
- Take a new photo with just one person
- If you already have a group photo with both of you, crop it into two separate photos before uploading—one of each person, even if it's a tight crop
Why it matters: If the model can't isolate one face, it tries to blend both faces into one, and you get identity bleed in the wide shot and face-off.
✓ Front-facing, both eyes visible, mouth unobstructed
What passes:
- Face pointing at the camera (0 to 15 degrees off-center is fine)
- Both eyes clearly open and visible
- Mouth unobstructed (no hand, no hair, no scarf)
- Chin visible and defined
What fails:
- Profile shot or 45-degree angle (the model treats it as looking away)
- One eye closed or mostly hidden by hair
- Mouth partially covered by a scarf, a hand, or a large beard
- Lips already open or mid-word (caught mid-laugh or mid-sentence)
How to fix:
- Stand facing the camera directly
- Brush hair out of your face if it's covering your eyes
- Close your mouth gently in a neutral or slight-smile position
- If you have a beard, make sure the lower lip is still visible
Why it matters: Lip-sync depends on the model seeing both your eye line and your mouth clearly. If the mouth is hidden, the model has to invent the mouth movement and it usually gets stuck or stops moving mid-word. If the eyes are hidden, the model can't lock onto your identity across the four shots.
✓ Natural light, no hard shadows on the face
What passes:
- Daylight from a window (best)
- Overcast daylight (soft and shadowless)
- Light coming from in front of you (not backlighting)
- Even lighting on both sides of the face
What fails:
- Hard shadows across half the face (sunlight through a gap, harsh overhead light)
- One side of the face in shadow, the other in light
- Backlit silhouette with the face dark
- Very warm tungsten light (gives a yellow cast that doesn't carry through the render)
- Overhead office lights in a dark room (flattens the face)
How to fix:
- Stand facing a window and take the photo with daylight on your face
- If you only have overhead lights, stand near a wall so the light bounces back and fills the shadows
- Take the photo at golden hour (sunrise or sunset) if you want warm light; the model carries that through better than tungsten
Why it matters: If the lighting is harsh or uneven, the model inherits that flatness or shadow into the render. A rendered shadow on your face stays as a rendered shadow in the video, and it usually looks like a rendering glitch, not intentional lighting.
✓ Sunglasses off, no masks, no hats pulled low
What passes:
- Bare face, eyes visible
- Hat off or worn back (not covering the eyes)
- Hair swept back or to the side (not covering the eyes)
What fails:
- Sunglasses (even reflective ones—the model can't read your eyes)
- Masks covering the nose or mouth
- Scarves wrapped around the lower face
- Beanies or baseball caps pulled down over the eyes
- Hair covering the eyes
How to fix:
- Take off sunglasses, masks, hats—all of them
- Brush hair back or to one side so both eyes are visible
- If you have a beard, trim or comb it so the mouth outline is still clear
Why it matters: Eyes and mouth are what the model uses to lock onto your identity and to generate mouth sync. If either is hidden, the model has to guess, and guessing usually results in distorted or static features. Sunglasses also make it impossible for the model to track expressions.
✓ Skip heavy filters, no upscaled screenshots
What passes:
- Raw phone photo, no beauty filter
- Slight saturation boost is fine
- A slight blur or vignette is okay (adds character)
- Original resolution your camera took it at
What fails:
- Heavy smoothing filter (your skin looks airbrushed)
- AI-upscaled or AI-enhanced photos (the model sees artifacts, not real faces)
- Screenshots blown up to fill the frame (pixelated)
- Extreme color grading (very saturated or very desaturated)
How to fix:
- Turn off beauty filters on your phone before taking the photo
- Use the raw phone image, not an edited version
- If you upscaled a photo, take a new one instead at native resolution
- Light editing (adjust brightness, contrast, crop) is fine; heavy smoothing is not
Why it matters: A heavily filtered or upscaled photo has fewer actual face details for the model to work with. A smooth, airbrushed face leaves nothing for the model to animate—it has no texture, no pores, no character. The render then interpolates between almost nothing and usually produces a blurry or distorted result. A raw photo has all the detail the model needs.
✓ Chest-up or waist-up framing
What passes:
- Both shoulders clearly visible
- At least the top half of your outfit visible (shirt, jacket, etc.)
- Chest and neck visible
- Arms in or out of frame is both okay
What fails:
- Just your head, no shoulders (too tight, the model doesn't understand the body)
- Head at the very edge of the frame, off-center
- Arms cropped off unnaturally (looks like a float)
- Only the face visible in a small portion of the frame
How to fix:
- Frame yourself so both shoulders and the top of your outfit are visible
- Leave some space above your head (not cropped at the hairline)
- Include your arms if you can, or let them be out of frame—both work
Why it matters: The model uses the shoulder line, the chest, and the outfit to understand your full body shape and movement. If you give it only a head, it has to invent the body underneath, and the proportions often look off. The outfit you wear in the photo is the outfit in the video, so the model needs enough framing to understand what you're wearing.
✓ Same lighting on both photos
What passes:
- Both photos taken in the same light (same window, same time of day)
- Both photos have similar brightness and color temperature
- Even if the two people are in different locations, the light feels the same
What fails:
- Photo 1 taken indoors at night, Photo 2 taken outside in bright sun
- One photo very warm/yellow, the other very cool/blue
- One photo very bright, the other very dark
- Lighting that makes the two faces look like they are in different rooms
How to fix:
- Retake both photos in the same location at the same time
- Or retake one of them to match the lighting of the other
- If the photos are from different days, try to match the time of day and type of light
Why it matters: If the lighting doesn't match, the render treats the two people as if they are in different rooms. The four-shot duet suddenly looks like a split-screen call instead of two people standing at the same mic in the same teal studio. Mismatched lighting breaks the illusion.
✓ File format and size
What passes:
- JPG, PNG, or WEBP format
- 20 MB or under
- 1080p resolution or native phone camera resolution is plenty
- Original dimensions (no extreme aspect ratios)
What fails:
- BMP or TIFF (not supported)
- 50+ MB (rejected on upload)
- Upscaled to 4K or beyond (wastes compute, no benefit)
- An extremely wide or tall crop (aspect ratio above 3:1)
How to fix:
- Export your photo as JPG (best) or PNG (if you need transparency, which you don't)
- Keep it at your camera's native resolution or crop it to a reasonable aspect ratio
- If it's too large, re-export at lower compression (JPG quality 80–90)
Why it matters: The upload system rejects oversized files and unsupported formats. Your phone camera's resolution is already optimized for video AI; upscaling beyond that wastes processing power and usually introduces artifacts.
The five common failures and how to fix them
Failure: The mouth does not move or moves wrong
Symptom: Lips barely open, or they freeze mid-word, or they move on the wrong syllables.
Root causes:
- The photo had the mouth already open or caught mid-laugh
- The mouth was partially hidden or blurred in the photo
- The uploaded audio had heavy processing, autotune, or overlapping vocals
Fix (in order):
- Take a new photo with the mouth closed in a neutral or slight-smile position
- Make sure the lower lip is clearly visible and sharp in the photo
- If you're using custom audio, use a clean vocal track with no effects
- Render again with the new photo
Failure: The face is blurry or looks like AI slop
Symptom: The face is soft, pixelated, or distorted, especially around the mouth and eyes.
Root causes:
- The original photo was blurry or out of focus
- The photo was heavily filtered or smoothed (beauty filter)
- The lighting was too dim or too harsh, creating shadows the model inherited
Fix:
- Retake the photo in sharper focus and brighter light
- Turn off all filters
- Use daylight instead of artificial light
- Make sure the face fills at least 40% of the frame so the model has enough detail
Failure: One face is much bigger or smaller than the other
Symptom: In the wide shot, one person looks far away and the other looks close.
Root causes:
- The two photos were framed at different distances (one close, one far)
- One photo has the face very small in the frame
Fix:
- Retake one or both photos so the faces are framed at similar distances
- Make sure both photos are chest-up or waist-up, not extreme close-ups or wide shots
Failure: The two faces blend into one or look identical
Symptom: In the wide shot or face-off, the two faces drift together and look like one person with two hairstyles.
Root causes:
- The two people look very similar (same age, hair length, hair color)
- Both photos are framed identically and lit the same way
- Both photos are taken in the same location with the same background
Fix:
- Take new photos with clearly different framing—one tight chest-up, one looser waist-up
- Take the photos in different locations so the backgrounds differ (even if the background won't appear in the render, it changes the feel of the photo)
- Wear distinctly different outfits in each photo
- Retake with one person in brighter light and the other in softer light (as long as both are still well-lit)
Failure: The upload is rejected
Symptom: Error message on upload, file not accepted.
Root causes:
- Wrong file format (BMP, TIFF, GIF instead of JPG/PNG/WEBP)
- File too large (over 20 MB)
- Corrupted file that won't open
Fix:
- Export as JPG or PNG from your phone or computer
- Check the file size and re-export at lower quality if needed (JPG quality 75–85)
- Take a new photo and try again
Before you upload to 58th Floor
Single-person photos only. If you have a group photo you want to use, crop it into two separate photos first.
Test one photo first. Upload one photo alone, render a test clip to see how it looks in motion. If the face looks good, upload the second photo.
Lighting matters more than background. Your background won't appear in the render (it's replaced with the teal studio), but your lighting will. Harsh shadows in your photo become harsh shadows in the video.
If the render looks bad, fix the photo, not the settings. The issue is almost always the photo quality or framing, not the resolution or length settings. Rerender with a better photo, not a higher resolution.
Frequently asked questions
Can I use a photo with someone in the background?
No. One person per photo. If there's someone else partially visible or blurred behind you, crop them out or retake the photo.
Can I use a screenshot from a video?
Not ideal. Screenshots are usually compressed, pixelated, and lower resolution than a phone photo. A fresh phone photo is always better than a screenshot.
Can I use a professional headshot?
Yes, if it passes the checklist. Professional lighting and makeup are fine; heavy filters and extreme smoothing are not. A natural-looking professional headshot works great.
What if I only have photos of me from far away?
Crop the photo so your face fills most of the frame, or retake a closer selfie. The model needs to see face detail to lock onto your identity.
Can I use a photo where I'm not smiling?
Yes. Neutral expression works fine. The model just needs the mouth to be unobstructed and clearly defined. A smile, a neutral look, or a slight frown all work.
What if one photo is from years ago?
It works, as long as it passes the checklist. The photo does not need to be recent. Many people use a favorite photo from years back. The model just needs to read the face clearly.
Can I use photos of pets?
Yes. One pet per photo, clear face in frame, natural lighting. Cats and dogs work especially well if the face is facing the camera.
What's the difference between good lighting and bad lighting?
Good: Daylight from a window, soft and shadowless. Bad: Overhead office light that flattens the face, harsh sunlight creating strong shadows, or backlit silhouettes. When in doubt, use window light.
Upload and render
Now that your photos pass the checklist, go to the homepage and upload them. If the render looks good on the first try, you're done. If it doesn't, come back to this checklist and pick the photo that best matches the failing symptom, then retake and re-render.
Read the full step-by-step guide if you want screenshots and more troubleshooting tips.