Turn One Photo Into a Moving Video with AI in 10 Minutes

Yes, you can turn a photo into a video with AI in about 10 minutes, without owning a GPU and without a monthly subscription. The recipe: rent an NVIDIA RTX 4090 on Glows.ai (from $0.49/hour, billed per second), open the preconfigured ComfyUI image, load the built-in Wan 2.2 image-to-video template, and upload your photo. The Wan team's own benchmark puts a 5-second 720p clip at under 9 minutes of render time on a single RTX 4090 — which works out to roughly $0.07 of compute per clip at Glows.ai rates.

This walkthrough covers:

• What you need before you start (a browser and about $0.50)
• Why an open model beats free online generators for anything past clip number six
• The four steps, with an honest minute-by-minute timeline
• What one clip actually costs, with sources you can check

What You Need Before You Start

• A photo. Portrait, landscape, product shot, pet — anything. Higher resolution helps; the model works from your image as the first frame.
• A Glows.ai account. Sign-up takes a minute at glows.ai. Promotional credits are sometimes available for new users — check the site or the Glows.ai Discord for current offers.
• About $0.50. One hour of RTX 4090 time costs $0.49 at current Glows.ai pricing (checked July 10, 2026), and billing is per second, so a 15-minute session costs about $0.12.
• No GPU, no installs. Everything runs in the browser on the rented machine.

Why an Open Model Instead of a Free Online Generator

Hosted image to video AI tools (Canva, Firefly, Vidu, Kling and friends) are genuinely the fastest option for one or two clips. The trade-offs appear at clip three: daily credit caps, watermarks on free tiers, queues, and content filters you can't adjust. Kling's free tier, for example, grants 66 credits per day — roughly six 5-second standard clips (AI Video Bootcamp's 2026 free-tools guide) — and typical paid plans across hosted generators run $20–100/month (imagine.art pricing survey, 2026).

The open alternative is Wan 2.2, an Apache 2.0-licensed video model from the Wan team. Its TI2V-5B variant generates 720p video at 24fps and is specifically built to run on a single consumer GPU like the RTX 4090. You keep full control: no watermark, no daily cap, your photo never enters a third-party generator's training pipeline, and the license permits commercial use.

At $0.49/hour, you'd need roughly 3,200 rental hours — about 6 clips a day for a year and a half — before buying the card wins on price.

Step 1: Create a ComfyUI Instance on Glows.ai (~2 minutes)

Log in to Glows.ai and open the Create New page. Set Workload Type to Inference GPU -- 4090 and select the ComfyUI image — it comes with ComfyUI preinstalled and serving on port 8188. For detailed screenshots of this flow, see the instance creation guide or our tutorial on running custom ComfyUI workflows on Glows.ai.

Click Complete Checkout. The instance boots in 30–60 seconds, and the My Instances page then shows three access links:

• HTTP Port 8188 — the ComfyUI interface (the one we want)
• HTTP Port 8888 — JupyterLab
• SSH Port 22 — terminal access

Note: Billing runs per second from boot to shutdown. There is no idle fee once you stop the instance, so there's no reason to keep it running between sessions.

Step 2: Load the Wan 2.2 Image-to-Video Template (~2 minutes, plus one-time downloads)

Click the Open link under HTTP Port 8188. In ComfyUI's template browser, pick the video section and select the official Wan 2.2 5B workflow — ComfyUI ships this as a built-in template.

The first time you load it, ComfyUI lists the model files the workflow needs — the 5B diffusion model, a VAE, and a text encoder, more than 15 GB combined — and offers to download them. Accept and wait. On the datacenter connection this takes minutes; on a home connection it's often the longest part of any local tutorial, which is exactly why we're not running this locally.

Reminder: If the template list doesn't show Wan 2.2, your ComfyUI version predates it. Update ComfyUI via the Manager, or drag the workflow JSON from the official ComfyUI docs straight into the canvas.

Step 3: Upload Your Photo and Describe the Motion (~1 minute)

This is the step that actually turns your photo into a video: in the workflow's Load Image node, upload the photo — it becomes the first frame. Then write a short motion prompt in the positive prompt box. Describe what should move, not what the image contains; the model already sees the image:

gentle breeze, petals drifting, slow push-in camera, soft natural light

Two settings worth knowing before your first run:

• Length: the template defaults to 121 frames (about 5 seconds at 24fps). Dropping to 49 frames (~2 seconds) cuts render time roughly proportionally — a good first-run choice.
• Resolution: the default 1280×704 is what the under-9-minute benchmark refers to. Lower it for faster drafts.

Step 4: Hit Run and Let the GPU Work (~4–9 minutes)

Click Run. The progress bar crawls through the diffusion steps, and the finished clip appears in the output node — download it as a standard video file. A short 49-frame draft lands in a few minutes; a full 5-second 720p clip takes under 9 minutes on the RTX 4090, per the Wan team's published numbers.

While it renders, queue variations: same photo, different motion prompts. Each queued job runs back-to-back, so an hour-long session ($0.49) yields around six full-length clips.

The Honest Timeline

"10 minutes" claims deserve an audit, so here is ours:

The honest print: a short 49-frame first clip fits inside 10 minutes; a full 5-second 720p clip on a first run lands closer to 12–15 minutes because of the one-time model download. Every clip after that is pure render time — under 9 minutes each — and your own hands-on time totals about 4 minutes. Store the models in Datadrive and even the download disappears from future sessions (see how to set a persistent ComfyUI model path).

What One Clip Actually Costs

Turning a photo into a video with AI this way is checkable from public pricing — no private benchmark required:

Even after padding each clip with setup and retries — call it 15 minutes of billed time per keeper — you're at roughly $0.12 per clip, or $3.70 for 30 clips a month. That's about a tenth of the entry-level hosted subscription, with no watermark and no cap.

Getting Better Motion From a Single Photo

• Prompt the motion, not the scene. "Waves rolling in, hair moving in wind, handheld camera drift" beats re-describing the photo.
• Add a negative prompt for the classics: distortion, morphing faces, extra limbs.
• Iterate cheap, then commit. Draft at 49 frames, and only render 121-frame finals for prompts that worked.
• Reuse your downloads. Mount Datadrive and point ComfyUI's model path at it so the 15+ GB download happens exactly once — the Hugging Face model download guide covers the fastest way to pull extra models.

FAQ

Do I need my own GPU to turn a photo into a video with AI? No. A rented RTX 4090 at $0.49/hour handles Wan 2.2's 5B image-to-video model comfortably — it has the 24 GB of VRAM the official pipeline targets, and per-second billing means a single session costs cents, not a monthly fee.

How much does one AI-generated clip cost this way? About $0.07 in compute for a 5-second 720p clip: under 9 minutes of render time (Wan team's number) at $0.49/hour (Glows.ai pricing, July 2026). With generous padding for retries, budget ~$0.12 per finished clip.

Is Wan 2.2 free for commercial use? Yes — Wan 2.2 is released under the Apache 2.0 license (GitHub), which permits commercial use. Check the individual licenses of any community LoRAs or checkpoints you add on top.

Why not just use a free online image to video generator? For one or two clips, do. Free tiers cap out fast — Kling's grants about six standard 5-second clips per day — and usually watermark output. Past that point, ~$0.07/clip on a rented GPU is cheaper than any subscription, and you control resolution, length, and content.

Does Glows.ai offer free credits? Promotional credits may be available depending on current campaigns — we don't list a number here because offers change. Check glows.ai or the community Discord for what's live right now.

Make Your First Photo Move Today

One photo, four steps, about $0.07 of compute. Create a Glows.ai account, launch a ComfyUI instance on an RTX 4090, and your first moving clip can be rendering within 10 minutes — stop the instance when you're done and the meter stops with it.