Check Reference Image Resolution Before You Spend Credits

Reference images are the backbone of identity-preserving AI video generation. They anchor faces, products, logos, and scene composition so the model can reproduce them across frames. But if your reference is too small or riddled with compression artifacts, the model interpolates missing detail—and the result is soft textures, melted edges, and uncanny distortions that no prompt tweak can fix. The cheapest fix is the one you make before hitting generate: check your reference image resolution.

**Why 1024px on the longest side?** Most current video generation pipelines internally resize references to a working resolution between 768px and 1024px. If your source is already below that threshold, the upscaler inside the pipeline invents pixels, introducing blur and hallucinated detail. Starting at 1024px or above ensures the model receives clean, information-dense input. For product shots with fine text or thin logo lines, aim for 1536px or higher to preserve legibility after the model's internal downsampling.

**The 3-step pre-flight checklist:**

1. **Check pixel dimensions.** Right-click the file and inspect properties (Windows), use Preview > Tools > Show Inspector (macOS), or run `identify image.png` in ImageMagick. Confirm the longest side is at least 1024px. If you are batch-processing references, a quick script using `exiftool` or Python Pillow can flag undersized files before they enter your workflow.

2. **Zoom to 100% and inspect for artifacts.** JPEG compression blocks, color banding in gradients, and ringing around high-contrast edges all degrade generation quality. Pay special attention to logo text, product labels, and skin tones. If you see blocky squares or smeared transitions at 100% zoom, re-export from the original source at higher quality or switch to a lossless PNG.

3. **Prefer PNG or high-quality JPEG (quality 90+).** Lossy formats discard high-frequency detail that the generation model relies on for sharp edges. PNG preserves everything. If file size is a concern, a JPEG at quality 92-95 is usually indistinguishable from lossless for photographic references while remaining significantly smaller than PNG.

**Common traps in team workflows.** Marketing teams often pull product images from web pages or compressed PDF decks. These sources are typically 72 DPI and 600-800px wide—well below the safe threshold. Social media downloads (Instagram, WeChat, LINE) apply aggressive recompression that destroys fine detail even when pixel dimensions look adequate. Always trace back to the original asset export from your design tool or DAM system.

**What to do when you only have a low-res reference.** If the original high-res file is genuinely unavailable, use a dedicated AI upscaler (Real-ESRGAN, Topaz Gigapixel, or similar) to bring the image to 1024px+ before feeding it to the video generator. Upscaling as a separate step gives you a chance to inspect the result and fix artifacts manually, rather than letting the video model hallucinate corrections across dozens of frames. This two-stage approach consistently produces sharper output and saves credits you would otherwise spend on regeneration attempts.

Making this 30-second resolution check part of your standard pre-generation workflow eliminates the most common cause of "why does my output look blurry" tickets. Your credits, your timeline, and your final export quality will all thank you.