ShadowReel Stealth Levels: Standard vs Enhanced vs Max Stealth
ShadowReel offers three stealth levels that control the intensity of modifications applied during content uniquification: Standard, Enhanced, and Max Stealth. Each level represents a different balance between visual fidelity and detection evasion strength. Standard applies the lightest modifications for platforms with basic detection. Enhanced targets machine learning classifiers used by Instagram and TikTok. Max Stealth applies the strongest modifications to defeat YouTube's Content ID and the most aggressive detection systems.
How Do the Three Stealth Levels Compare?
Every stealth level modifies the same set of parameters, but at different intensities. The table below provides a complete technical comparison across all modification dimensions, including the resulting visual impact and recommended use cases.
| Parameter | Standard | Enhanced | Max Stealth |
|---|---|---|---|
| SSIM (quality score) | >0.97 | >0.92 | >0.85 |
| Noise Sigma | 1.5 | 3.0 | 4.0 |
| Color Shift | ±2 | ±6 | ±8 |
| Crop | 1px | 3px | 2% |
| Rotation | ±0.15° | ±0.5° | ±0.7° |
| Horizontal Flip | No | No | 50% chance |
| Vignette | None | 12% | 15% |
| Video CRF | 17 | 18 | 19 |
| Visual Impact | Imperceptible | Minimal | Slight (visible on close inspection) |
| Best For | Twitter/X, Reddit, Pinterest | Instagram, TikTok, Facebook | YouTube, Content ID systems |
SSIM (Structural Similarity Index) measures the visual similarity between the original and processed file on a scale of 0 to 1, where 1 is identical. A score above 0.95 is generally considered imperceptible to the human eye. CRF (Constant Rate Factor) controls video compression quality on a scale of 0 to 51, where lower values mean higher quality. The range of 17-19 used by ShadowReel produces visually lossless to near-lossless output.
When Should You Use Standard Stealth?
Standard stealth is the right choice when you need to bypass basic duplicate detection with zero visible changes to your content. At SSIM greater than 0.97, the output is structurally 97% identical to the input. The modifications are confined to noise sigma 1.5 (barely above sensor noise in most cameras), color shifts of plus or minus 2 (invisible to the human eye), 1-pixel crop, and 0.15-degree rotation. No vignette or flip is applied.
Standard stealth reliably defeats file hashing and perceptual hashing (pHash/dHash). It is sufficient for platforms that rely primarily on these two methods: Twitter/X, Reddit, Pinterest, and most forums or content management systems. These platforms do not currently deploy ML classifiers or Content ID for general user uploads.
Use Standard when quality is the top priority and the target platform does not use advanced ML-based detection. It is also the fastest processing option, as the lighter modifications require less computation.
When Should You Use Enhanced Stealth?
Enhanced stealth is designed for platforms that use machine learning classifiers alongside perceptual hashing. Instagram and TikTok both deploy neural network models to generate embedding vectors for uploaded content, and these models are more resistant to simple modifications than perceptual hashing alone. Enhanced stealth increases modification intensity across every parameter to shift the ML embedding vector beyond the similarity threshold.
At SSIM greater than 0.92, the output retains 92% structural similarity. The differences are minimal and only detectable when comparing the original and processed files side-by-side at full resolution. In normal viewing conditions, scrolling through a feed or watching in a Stories viewer, Enhanced stealth output is indistinguishable from the original.
The key additions over Standard are: noise sigma 3.0 (noticeable only in smooth gradient areas at 100% zoom), color shifts of plus or minus 6 (within the range of normal display calibration variance), 3-pixel crop, 0.5-degree rotation, and a 12% vignette that subtly darkens the edges. These modifications collectively alter enough dimensions in the neural embedding to defeat ML classifiers.
When Should You Use Max Stealth?
Max Stealth is the most aggressive level and is specifically engineered to defeat YouTube's Content ID system, which analyzes both audio and visual fingerprints at the segment level. Content ID is the hardest detection system to bypass because it can match partial segments of your video against a reference database of over 100 million files. Max Stealth applies the strongest modifications across all parameters to ensure no segment retains a matchable fingerprint.
At SSIM greater than 0.85, the output retains 85% structural similarity. This is still high enough that the content looks good in normal viewing conditions, but on close side-by-side inspection, some changes may be noticeable: a slight warmth or coolness to the color palette, faint grain in smooth areas, and subtle vignetting at the edges. The 50% chance of horizontal flipping (when enabled) completely changes the geometric orientation of the content.
Max Stealth uses noise sigma 4.0, color shifts of plus or minus 8, 2% crop (removing content from all edges), 0.7-degree rotation, 15% vignette, CRF 19 for video, and a 50% chance of horizontal flipping. The audio track also receives the strongest EQ, resampling, and tempo modifications.
What Is the Quality vs Evasion Tradeoff?
There is an inherent tradeoff between visual quality and detection evasion strength. Stronger modifications push the digital fingerprint further from the original, which improves evasion, but they also introduce more visible changes. ShadowReel's three levels are calibrated to provide the minimum modification necessary for each class of detection system.
- Standard (SSIM >0.97): Zero visible difference. Defeats hash-based detection. Does not reliably defeat ML classifiers.
- Enhanced (SSIM >0.92): Differences visible only at 100% zoom in side-by-side comparison. Defeats hash-based detection and ML classifiers.
- Max Stealth (SSIM >0.85): Slight changes visible on close inspection. Defeats all four detection layers including Content ID.
The recommendation is to always use the lowest stealth level that reliably defeats the target platform's detection. Using Max Stealth for a Twitter/X post applies unnecessary modification. Using Standard for a YouTube upload risks a Content ID claim. Match the stealth level to the platform.
Are There Any Caveats With Max Stealth?
Max Stealth may produce visible changes on close inspection. Always test before posting. The 50% horizontal flip chance can be problematic for content that contains readable text or recognizable asymmetric elements (logos, text overlays, branded graphics). If your content includes text, disable the flip option or use Enhanced stealth instead.
The 2% crop removes a small border from all edges, which may cut off content positioned at the extreme edges of the frame. For content composed tightly to the edges, consider using Enhanced stealth or ensuring your source material has adequate safe margins.
CRF 19 produces slightly more compression artifacts than CRF 17, particularly in areas with fine detail or high-frequency textures. For source material that is already heavily compressed, the additional generation loss may be noticeable. When possible, start from the highest quality source file available.
Which Stealth Level Should You Use for Each Platform?
The table below provides specific recommendations for each major platform based on the detection methods they employ. These recommendations balance the minimum modification needed with a safety margin for algorithm updates.
| Platform | Recommended Level | Why |
|---|---|---|
| YouTube | Max Stealth | Content ID uses segment-level audio + visual fingerprinting, the most aggressive system |
| Instagram Reels | Enhanced | ML classifiers generate neural embeddings; perceptual hashing alone is insufficient |
| Instagram Feed | Enhanced | Same ML classifier pipeline as Reels, applied to both images and videos |
| TikTok | Enhanced | ML-based near-duplicate detection on all uploads; aggressive reach reduction on matches |
| Enhanced | Rights Manager + perceptual hashing; Enhanced covers both detection layers | |
| Twitter/X | Standard | Relies on file hashing and basic perceptual hashing; no ML classifiers for general uploads |
| Standard | File hashing and perceptual hashing only; lightweight detection system | |
| Standard | Perceptual hashing for visual search; Standard is sufficient to alter the hash | |
| Twitch | Max Stealth | Uses Audible Magic audio fingerprinting similar to Content ID; audio modifications required |
If you are unsure which level to use, Enhanced is the safest general-purpose choice. It defeats three out of four detection layers and maintains high visual quality. Only step up to Max Stealth when targeting YouTube or Twitch, and only step down to Standard when you are certain the platform uses basic detection only.