Correction, 2026-10-02: this page said cheap models blur fast movement and that Kling v3 Standard "handles motion better than Wan at 24". We never measured either. The motion section is now a frame-by-frame check plus things to try, and it says that Kling v3 Standard needs a start image. The page named Clarity first among the upscalers, but the upscale tool opens on Topaz; it now lists all three, all for images only: Topaz (the default, a fixed 4x enlargement at 11 credits per started 24-megapixel block of output), Clarity (13 credits, always 2x on 8frame, diffusion-based) and BRIA (6 credits, 2x or 4x, up to 8192 x 8192). It said an upscaled start image "genuinely works" and that its sharpening "propagates into every frame", which we have not measured either. It described how platforms recompress uploads without a source; that section is now a comparison you can run yourself. We added the output resolution of Veo 3.1, Veo 3.1 Fast and Kling v3 Standard, a note that the full-resolution final is a new generation rather than the draft enlarged, and the Sources section the page was missing.
Blurry AI video usually comes from one of five places, and they need different fixes. Working out which one you have takes a few minutes and can save you from regenerating at a higher price for a problem that was never about the model.
Before the list, the constraint that shapes all of it: there is no video upscaler on 8frame, and no video model on 8frame outputs 4K. The upscale tool works on images only, so you cannot sharpen a finished clip there. Every fix below happens before or during generation.
TL;DR
- The resolution setting comes first: Wan 3.0 defaults to 1080p at 95 credits for 5 seconds, and 480p is 24. Check what the model actually outputs
- Motion: sharp single frames in a clip that looks soft usually means motion blur, which may be correct. Soft single frames point to resolution or the start image
- No video upscaler. Generate the final at the resolution you deliver
- Compression after upload: compare your downloaded file with the uploaded version
- A soft start image in image-to-video: check the still at 100% before you change the model
1. The resolution setting
Check this first. A 480p clip viewed at full screen on a 1080p display looks soft on any model.
8frame prices and output resolutions, a credit being about $0.01 at pack rate:
| Model | Clip | Credits | Output resolution |
|---|---|---|---|
| Wan 3.0 | 5 s | 24 / 48 / 95 | 480p / 720p / 1080p (default 1080p) |
| Wan 3.0 Prime | 5 s | 46 / 95 / 189 | 480p / 720p / 1080p (default 1080p) |
| Seedance 2.5 | 5 s | 70 / 157 | 480p / 720p (default 720p) |
| Veo 3.1 Lite | 8 s with audio | 54 | 720p |
| Kling v3 Standard | 5 s, with sound or silent | 86 / 57 | 720p |
| Hailuo H3 | 5 s | 88 | 2K |
| Veo 3.1 Fast | 8 s with audio | 168 | 1080p |
| Veo 3.1 Standard | 8 s with audio | 448 | 1080p |
The fix: set the resolution before you generate. Base Wan 3.0 at 1080p is 95 credits for 5 seconds, the cheapest 1080p Wan; Prime costs about twice as much at every resolution. Some models have no setting: Kling v3 Standard returns 720p and Veo 3.1 Lite is 720p, so if you deliver 1080p, pick a model that outputs it. The full list is in AI video resolution by model.
2. Motion
What to check: step through the clip frame by frame. If single frames are sharp and the clip looks soft only while it plays, you are looking at motion blur, which may be correct: a real camera blurs fast movement too (see what motion blur is). If single frames are soft, go back to the resolution (section 1) or the start image (section 5).
What to try: slow the motion in the prompt ("slow, deliberate movement") and run it against your original prompt on the same model. If the blur stays, run the same prompt or start image on a second model at the same resolution and compare the same frames. At 720p, Wan 3.0 is 48 credits for 5 seconds, Kling v3 Standard is 57 without sound and needs a start image, and Seedance 2.5 is 157. We have not measured which model keeps fast motion sharpest, so judge on your shot.
3. Expecting an upscale that is not there
This is the expectation problem, so here it is plainly: 8frame upscales still images, not video. The upscale tool opens on Topaz, and all three of its options output images:
- Topaz (the default): a fixed 4x enlargement, 11 credits per started 24-megapixel block of output, up to 281
- Clarity: 13 credits, always 2x on 8frame. It is diffusion-based, so it redraws detail rather than only enlarging it
- BRIA Increase Resolution: 6 credits, 2x or 4x, up to 8192 x 8192
There is no "fix it afterwards" step for video. If you drafted at 480p intending to upscale the keeper, that plan does not work here.
The fix: decide the delivery resolution before you generate the final. Draft at 480p for 24 credits, then generate the final at the resolution you deliver. Three drafts plus one 1080p final on base Wan 3.0 is 167 credits, about $1.67. The 1080p run is a new generation, not the draft enlarged, so it will not match the draft frame for frame: keep the prompt and start image fixed and judge the final on its own.
4. Compression after upload
Your file can be sharp and the upload can look soft. 8frame does not re-encode video: the file you download is the one the provider returned, H.264 in MP4 at 24 fps on every model we measured. Whatever happens after you upload is the platform's processing, not your file.
What to check: compare your downloaded file with the uploaded version on the same screen. If your file is sharp and the upload is soft, the softness was added after upload, and regenerating will not change it.
What to try: export to the platform's own recommended upload settings rather than something larger. When you compare, look closely at heavy grain, very fine detail and fast full-frame motion in both versions.
5. A soft start image
In image-to-video, the model starts from your still, so check the still before you change the model.
What to check: open the start image at 100%. If it is smaller than the clip you are making (a 16:9 1080p clip is 1920 x 1080 pixels), it is a candidate for an upscale. If it is already that size or larger and still looks soft, the softness is in the image itself, and regenerating the still may serve you better than enlarging it.
What to try: this is the one place the upscale tool fits a video workflow, because it works on stills. Topaz (the default) enlarges 4x from 11 credits; BRIA enlarges 2x or 4x for 6; Clarity enlarges 2x for 13, and because it is diffusion-based, compare faces, text and logos against your original before you animate. We have not measured how much an upscaled still changes the finished clip, so run the same model on the original and on the upscaled still and compare.
The diagnostic order
- Check the resolution setting and what the model outputs.
- Look at one frame. Sharp frame and soft motion means motion blur. Soft frame means resolution or start image.
- If image-to-video, check your start image at 100%.
- Compare the downloaded file with the uploaded version. A difference means it happened after upload.
- If all of the above check out, run the same prompt or start image on a second model at the same resolution and compare.
FAQ
Can I upscale AI video on 8frame? No. The upscale tool works on images only (Topaz, Clarity and BRIA) and opens on Topaz. Generate the final at the resolution you deliver; no video model on 8frame outputs 4K.
Why does Wan 3.0 cost 95 credits sometimes and 24 other times? Resolution. It defaults to 1080p, its most expensive setting: 95 credits for 5 seconds, against 48 at 720p and 24 at 480p. Wan 3.0 Prime has the same default and costs about twice as much at every resolution.
Is 480p ever enough? For drafts, yes: you are checking motion and composition, not detail. For delivery, put the 480p clip in your real timeline at full screen and judge it there before you decide.
My frames are sharp but the video looks soft. Why? That is likely motion blur, which may be correct. If you want it crisper, slow the motion in the prompt, or compare a second model at the same resolution.
What resolution does each model output on 8frame? Seedance 2.5 and 2.0 offer 480p or 720p; Kling v3 Standard returns 720p; Veo 3.1 and Veo 3.1 Fast are 1080p and Veo 3.1 Lite is 720p; Wan 3.0 offers 480p, 720p or 1080p; Hailuo H3 outputs 2K. No video model on 8frame outputs 4K.
Sources
- 8frame prices: 8frame tool registry credit functions calculateWan3Cost (Wan 3.0, 24 / 48 / 95 credits for 5 s at 480p / 720p / 1080p; Wan 3.0 Prime 46 / 95 / 189), calculateSeedance2Cost (Seedance 2.5, 70 at 480p and 157 at 720p for 5 s), calculateVeo31Cost (8 s with audio: Veo 3.1 Lite 54, Veo 3.1 Fast 168, Veo 3.1 Standard 448), calculateKlingV3Cost (Kling v3 Standard, 86 with sound and 57 silent for 5 s) and calculateHailuoH3Cost (Hailuo H3, 88 for 5 s), read 2026-10-02
- Default settings (Wan 3.0 and Wan 3.0 Prime at 1080p, Seedance 2.5 at 720p): 8frame front tool registry, read 2026-10-02
- Output resolutions per model, Kling v3 needs a start image, no video model outputs 4K: 8frame video generation code and front tool registry, via the 8frame content fact sheet, read 2026-10-02
- Upscale tool: 8frame front tool registry (image only, opens on Topaz, options Topaz, Clarity, BRIA 2x and BRIA 4x; BRIA up to 8192 x 8192) and backend credit function calculateTopazCost (fixed 4x, 11 credits per started 24-megapixel block of output, up to 281), read 2026-10-02; Clarity sent at a fixed scale factor of 2: 8frame image upscale code, via the 8frame content fact sheet, read 2026-10-02
- Clarity is diffusion-based: Clarity upscaler README, https://github.com/philz1337x/clarity-upscaler, checked 2026-10-02 ("free and open-source Magnific Alternative", built on a Stable Diffusion checkpoint with a ControlNet tile model and tiled diffusion)
- No re-encoding; H.264 in MP4 at 24 fps: generated video files on 8frame over the 120 days to 2026-09-30, measured per model (every measured model returned 24 fps); 8frame's media worker reads metadata only and serves the provider's file
Set the resolution before you generate, not after.
Try it without signing up: the 8frame canvas opens on a real board. The canvas is free and unlimited; generation is paid from $19/month.