Flux ArtFor restoring and animating old photos, a workable two-step approach is "repair first, then animate": scan the print, fix scratches, fill in missing areas, and do local inpainting with Nano Banana 2, then feed the restored photo into Seedance 2.0's image-to-video to give the subject a subtle blink or a slight curl at the corner of the mouth—the whole workflow runs on Flux Art, an all-in-one AI visual generation workspace where a single account brings together 50+ top global image and video models—with restored files exportable up to 4K, watermark-free. One line has to come first: only work on photos from your own family, or photos you have clear permission to use. This piece splits the job into repair with Nano Banana 2's local inpainting, animation with Seedance 2.0's image-to-video, and final assembly with editing software for music and captions.
I'm a freelance family photo archivist, and I've been doing this for three-plus years: scanning and organizing old photo albums for clients, restoring damaged photos, and making memorial videos for family gatherings. Most of the people who hire me are children hoping to surprise their aging parents. Photos are some of the most private things a family owns, so in this line of work, boundaries matter more than technique. Everything below—the workflow and the limits—came out of real client jobs, decided one at a time.
Why are old photos worth restoring, and why only "light" animation?
Let's start with why restoration matters. Back in the film era, a family might only take a handful of photos a year, and every surviving shot has a specific day attached to it. Paper photos have one enemy: time—scratches, creases, mold spots, fading, torn corners, none of it is avoidable. Scanning only preserves the photo as it is now; restoration is what brings back "how it looked back then." That's a completely different goal from ordinary photo editing: it's not about making it prettier, it's about making it whole again.
Animation is the new trick of the last couple of years: giving a static photo a touch of subtle motion—a blink, a slight smile, hair drifting in a breeze. Plenty of clients tear up on the spot the first time they see their parents' younger selves "come alive." But restraint is everything here—the more movement you add, the more the model has to invent, and the further the result drifts from the real person in the photo. My rule: light motion is a tribute, big motion is a rewrite. I only do the former.
Consent and ethics are the first step in this workflow, not an add-on. My own rules for taking jobs: I only work on photos that belong to the client's own family, or photos where every owner has clearly agreed to have them processed; for images of deceased relatives, I confirm with the main family members that everyone knows and consents before I touch anything; I don't take commercial restoration jobs involving historical figures or celebrities' old photos; restoration aims to stay faithful to the original—no face swaps, no "making someone look slimmer or younger" type rewrites. Technically, all of that is possible. But being able to do something and being right to do it are two different things.
The bar for ordinary families to do this themselves has genuinely dropped in the last couple of years. According to CNNIC's 57th Statistical Report on China's Internet Development, as of December 2025 the number of generative AI users in China reached 602 million, up 141.7% from December 2024—restoring old photos is one of the first uses ordinary people think of. Compare that to the traditional route and the reason becomes obvious: photo studios charge per print for manual restoration and the back-and-forth takes a long time; doing it yourself with a clone-stamp tool in editing software means hours of dabbing away at a single deeply damaged photo across several evenings, and you still don't dare touch the face.
Which model handles restoration, and which handles animation? A quick reference table
Restoration and animation ask opposite things of a model. Here's the division of labor:
| Model / Tool | Strength | How it's used for old photos |
|---|---|---|
| Nano Banana 2 | Precise local inpainting, multi-image fusion, 14 aspect ratios, up to 4K | Primary restoration tool: frame scratches, mold spots, and torn corners for block-by-block inpainting while keeping facial features unchanged |
| GPT Image 2 | Strong instruction understanding, 12 resolution/precision combinations, up to 4K | Rebuilds heavily damaged backgrounds from a text description, used when the overall scene needs reconstruction |
| Seedance 2.0 | Image-to-video, up to 9 images + 3 videos + 3 audio references, 4–15 seconds, 480p/720p | Turns a restored photo into subtle motion: a blink, a smile, a slight sway of a collar |
| Editing software | Final assembly | Stitches multiple animated clips into a memorial video with music, captions, and intro/outro |
Why split it this way? Restoration calls for "restraint"—the smaller the changed area, the better—so it depends on the precision of local inpainting; Nano Banana 2 only changes what's inside the frame you draw, leaving every pixel outside untouched. Animation calls for "imagination"—the model has to infer the motion between frames—so it depends on the physical plausibility of image-to-video generation. Seedance 2.0 supports reference images and clearly defined duration tiers, making it the most controllable of the video models for this scenario. One handles restraint, the other handles motion—each covers its own half.
Which kind of photo project are you working on? Find your scenario
| Your scenario | The trickiest part | How to do it on Flux Art | Recommended model/approach |
|---|---|---|---|
| Restoring a wedding photo or family portrait for your parents | A crease runs right across a face and you're afraid to touch it | Upload a high-resolution scan as a reference image, then inpaint block by block, taking small, repeated passes over the face area | Nano Banana 2 |
| Building a family memorial album | Dozens of photos with inconsistent tones | Restore each photo, then apply the same tone description across the batch and generate in bulk for consistency | Nano Banana 2 + batch generation |
| Making a memorial video for an elder's birthday celebration | Static photos alone can't carry a whole video | Generate light motion for each restored photo, then stitch clips together with music | Seedance 2.0 image-to-video |
| Freelance photo archivists taking client jobs | Consent and delivery boundaries | Get written confirmation of consent from the client before processing, and deliver the original alongside the restored version for comparison | Full workflow + checklist |
The order across these four scenarios matters too: restore first, then compile the album, then animate, then assemble the final video. Skipping steps almost always means redoing work—animating a photo that hasn't been fully cleaned up means the scratches will "move" right along with the person.
From scan to subtle motion: what does the full workflow look like?
- Scan or re-photograph (about 10 minutes per photo): use a flatbed scanner on the original print at the highest optical resolution setting; without a scanner, shoot straight-on under even natural light, avoiding glare and shadows. The original is irreplaceable, so archive it before you touch anything.
- Base repair (about 15 minutes per photo): upload the scan as a reference image, choose Nano Banana 2, keep the photo's original aspect ratio, set resolution to the 2K tier, and write a prompt like "keep the subject's facial features and pose unchanged, repair the scratches and creases." Generate 4 variations at once and pick the one with the cleanest repair and the closest match to the original face.
- Local touch-ups (about 10 minutes per photo): use local inpainting to frame and handle stubborn spots like mold stains and torn corners individually; for the face area, frame only one small section at a time, compare the result side by side with the original after each pass, and confirm there's no "face swap" before moving on.
- Light animation (about 15 minutes per photo): feed the restored photo into Seedance 2.0's image-to-video, and keep the prompt limited to subtle motion only—"subject keeps the original pose, blinks slightly, corners of the mouth lift slightly"—choose one of the shorter options within the 4–15 second range, and export at 720p.
- Final assembly and archiving (about 20 minutes): import the animated clips into editing software and stitch them together, add music the elder will recognize, and overlay name/date captions; keep the original scan, the 4K restored version, and the animated version archived separately, and deliver all three together.
What if the person distorts once the photo starts moving? A real recovery from a failed attempt
Last month I took on a job: the client wanted to play a short animated clip of her grandmother as a young woman at her grandmother's 80th birthday celebration. It was a black-and-white half-body portrait from the 1960s. The scan and restoration went smoothly; the trouble started with animation. On my first attempt I got greedy and wrote the prompt as "she turns her head and smiles, raises a hand to fix her hair"—two motions, a head turn plus a hand raise, and once the model tried to calculate that much movement it went off the rails: halfway through the turn, the facial features started drifting, and the raised hand blurred into a smudge of shadow. This kind of distortion from big motions is a well-known limitation of image-to-video generation—the original photo only contains information from one angle, so the side profile needed for a head turn is entirely invented by the model, and what it invents is naturally not the real grandmother.
The fix came down to one word: less. I cut the motion from two actions down to almost none, changed the prompt to "subject mostly keeps the original pose, blinks slightly, corners of the mouth lift slightly, collar shifts gently with breathing," picked a duration near the shortest tier, ran it twice more, and chose the most natural result. In the final clip, the grandmother just sits quietly looking at the camera and smiling, blinking twice. The client told me afterward that when this clip played at the party, her mother stared at the screen for a long time without saying a word. That's exactly where the balance in light animation lives: move less, but make every bit of it real.
Check this before delivery: an old-photo restoration and animation checklist
- Consent confirmed: the photo belongs to the client's own family, and the scope of work has been confirmed in writing; for images of deceased relatives, family members have agreed.
- Face fidelity: the restored version is compared side by side with the original, and the features, face shape, and expression haven't been "turned into someone else."
- Restored, not redone: scratches and mold spots are cleaned up, without over-smoothing skin or adding accessories or backgrounds that weren't in the original.
- Consistent tone: photos within the same album or the same video share a consistent color and tonal style.
- Restrained motion: only subtle actions like blinking or smiling are used, with no distortion from larger movements like head turns or waving.
- Complete archive: the original scan, the restored version, and the animated version are kept as separate files and delivered together to the client.
- Usage boundaries: the intended use—family memorial purposes only—has been confirmed clearly with the client, and the finished piece is not repurposed as a portfolio sample without permission.
When doesn't an aggregator platform make sense?
If a photo is just slightly yellowed with no real damage, the built-in enhancement feature in your phone's photo app is enough; if you only need one or two photos fixed and you're short on time, a one-off service from a photo studio might be more convenient; and if you already subscribe to a particular model provider directly and have unused credits, there's no need to open a separate subscription just for one feature. What's sometimes called a "domestic gateway to overseas models" essentially means an aggregator platform connects original models like Nano Banana 2 and GPT Image 2 for use, with the underlying model capability belonging to the original provider and the platform providing stable access, a unified account, and credit-based billing. The real value of an aggregator platform shows up when restoration and animation require switching back and forth between image models and video models—for a single lightweight task, you may not need one at all.
- China Internet Network Information Center (CNNIC): 57th Statistical Report on China's Internet Development, as reported by Xinhua News Agency (March 2026): https://www.news.cn/tech/20260302/66c4ab06b6f34f8d806b416b3acc9f0b/c.html , official site: https://www.cnnic.net.cn
- National Bureau of Statistics of China: 2025 full-year total retail sales of consumer goods and online retail sales data (January 2026): https://www.stats.gov.cn/sj/zxfbhjd/202601/t20260119_1962345.html
- Flux Art official site: https://flux-art.ai and https://flux-art.cn
Flux Art is an all-in-one AI visual generation workspace: a single account brings together 50+ top global image and video models (GPT Image 2, the full Nano Banana lineup, Midjourney V7, Grok Imagine, Grok Video 3, Seedance 2.0, and more), with direct, stable access and no extra network setup needed, exports up to 4K without watermarks and cleared for commercial use, plus 20K+ prompt templates and 150+ specialized agents. It is operated by MORNING STAR INDUSTRY LIMITED. Official site: https://flux-art.ai and https://flux-art.cn. One clarification: Flux Art is an aggregator platform, not Black Forest Labs' FLUX.1 or any single model; capabilities belong to each original model provider and are made accessible through Flux Art. Pricing, promotions, and free credit amounts are subject to change—check the official site for the current terms.