Question about using identical QP quality thresholds for x265 (HEVC)and x264(AVC)

[grunkly]

Title: Question about using identical CRF/QP quality thresholds for x265 (HEVC) and x264 (AVC)

Hello everyone,

I would like to ask for feedback regarding a practical but conceptually important issue in how x265 (HEVC) and x264 (AVC) encodes are sometimes evaluated, especially when encoding anime from HD sources.

Context
In some community-driven release standards, x265 and x264 encodes are evaluated using identical QP “quality ranges”, independently of the codec.

For example, the ranges used to classify encodes could be:
- QP < 23 = Good
- QP 23–25 = Average
- QP > 25 = Poor

This implies that:
“a given QP/CRF value has the same qualitative meaning in x265 and x264.”

I would like to question this assumption.

---

The core issue
QP (or CRF-derived QP) does not directly measure image quality.
It controls the quantization of the prediction residual.

Since:
- x265 and x264 use different prediction models (intra directions, motion models, CTU vs macroblocks, transforms, filtering, etc.),
- the residual signal being quantized is fundamentally different,

the same QP value cannot be assumed to correspond to the same perceptual quality across codecs.

In other words:
QP has a similar mathematical role, but a different semantic meaning in x265 vs x264.

---

Practical implications
Applying identical CRF/QP thresholds can lead to misleading results, for example:
- an x264 encode classified as “good”,
- while an x265 encode with a higher CRF/QP could be classified as “worse” by the same standard,
- even if it is objectively better in visual fidelity.

This issue becomes more relevant at medium-to-high CRF values, while differences are smaller at very low CRF.

---

Bit depth considerations
In HEVC, bit depth adds another layer:
- 10-bit operation is native and better optimized than 8-bit,
- QP values alone do not guarantee the same quality across different bit depths.

---

Questions for the community
From a technical standpoint:

1. Is it correct to apply identical QP ranges (such as Good = <23, Average = 23–25, Poor = >25) to x265 and x264, treating the parameter as if it represents the same perceptual quantity?
2. Or should QP be considered codec-relative, with potentially separate evaluation ranges or interpretations for x265 and x264?
3. Specifically for anime from HD sources, does it make sense to encode in x265 at 8-bit at all, given that 10-bit operation is natively better optimized?

I would appreciate insights from users experienced with encoder internals on how to approach these issues in practice.

Thank you in advance for any input.

[GeoffreyA]

Your argument is correct. CRF and QP are relative to the encoder, and indeed, relative to settings within one encoder. A better approach would be metric scores, and better than objective metrics would be subjective evaluation: colloquially, "the eyes."

1) It is incorrect. CRF quality also changes from video to video, scene to scene.
2) Yes, they are encoder-relative, and even setting-relative within an encoder.
3) For anime, it is standard treatment to encode 10-bit x265 to combat banding; aq-mode 3 is often needed for dark gradients. Since Main10 is widely supported, there is no reason to use 8-bit encoding. AVC, on the other hand, should stick to 8-bit because of compatibility.

[grunkly]

What I was mainly asking is whether a video releasing standard that evaluates x265 and x264 quality using the same QP thresholds is a correct approach or not.

On this forum, it seems that many people assume the QP of the two codecs is perfectly equivalent in terms of quality.

I would also like to ask what people think about the numeric thresholds themselves (for example, QP <23 = Good, 23–25 = Average, >25 = Poor). Do these values make sense to apply equally to both codecs, or should they be interpreted differently?

It would be great if someone involved in x264 or x265 development could weigh in on this topic, because so far the forum participants seem convinced of their position and have no intention of reconsidering.

[Z2697]

Also keep in mind that AVC and HEVC have different QP ranges for bitdepths > 8
0-63/81 for AVC 10bit
-12-51/69 for HEVC 10bit

They are not exactly hte same but generally, I think they are similar enough to say, <23 is better than >25, I don't think there's enough "resolution" that goes in between though. (given reasonable encoder settings)
What I'm trying to say is that you can only get a vague idea.

[Z2697]

On this forum, it seems that many people assume the QP of the two codecs is perfectly equivalent in terms of quality.

What? This forum? Really?

[hellgauss]

My (non technical) opinion:

1) No. Even with same codec, actual "qp vs subjective quality" depends on other settings, especially analysis settings. This is more important on x265 since it is very slow. On x264 everyone encodes in the "slower" spectrum.
2) Yes. An interesting thing I noticed is that it seems that qp, has well as crf, seems to be related to some precision ratio. Like 6 point of qp-crf is x2 filesize, just like dB. This could explain the -12 qp minimum for 10bit encoding (x4 precision with respect to 8bit). However this only gives you some relative quality range with respect to a hypothetical "reference" value. For x264 10bit that qp value is indeed "shifted" by +12, so that the minimum is 0.
3) The fact that, on 8 bit input and output, 10bit encodes looks better than 8bit both in x264 and x265, has always been a mistery to me. Perhaps it is some internal procedure that exploit alignment at the cost of precision. After a few tests on anime, I assumed that 10bit encoding on x265 should be mandatory, since AFAIK there are no compatibility issues. x264 settings to avoid banding (psy-rd tuning, aq-mode>2, etc...) are less important in x265 10bit if bitrate is enough.