♫ ByteBeats

The work and its context

ByteBeats makes an unusually direct connection between a numerical procedure and its sensory consequences. A compact expression supplies values that can become both sound samples and marks. The images do not illustrate a recording made elsewhere. They expose selected relationships within the procedure that generates it. In the four initial frames examined here, the results range from stark binary stripes to densely patterned orange fields and a pink waveform. Their differences concern how information is displayed as much as which information the underlying expression produces. ^{1 2 3 4 5}

Dadabots and kai released the 512-work project through Art Blocks Factory in April 2021. The official description identifies six families and four visualisation types, with the generated expressions described as occupying roughly forty to one hundred and forty bytes. That claim concerns the small synthesis expressions, not the complete program that selects parameters, manages audio and draws the screen. The inspected renderer is considerably longer. Keeping those scales separate makes the project's economy more intelligible: a short rule still needs an environment in which it can be experienced. ^{6 5}

This essay examines the program and four identifiable initial renderings. A subsequent browser check established changing visual output for #0 and the collection-held #472 after activation, alongside an advancing audio-context clock. It does not offer a listening review: the diagnostic clock is not evidence of what a listener hears. No claim about the heard character of an individual edition follows from its feature name. ^{5 1 2 3 4}

Conception and development

The collaborators' project site situates bytebeat beside Dadabots' experiments with much larger neural models. Its point of comparison is the amount of behaviour that a small mathematical expression can contain. The site credits Viznut with popularising the form and Kyle McDonald with bitmap visualisations. These are the collaborators' historical attributions; they are not evidence that this essay has reconstructed the complete history of either contribution. ⁷

The same site discusses ambitions for finding relationships between audio and compact expressions. An ambition described in project material should not become a claim that a finished learning system produced this edition. The inspected Art Blocks script instead contains explicit families, weighted choices and numerical transformations. The distinction locates the authorship where the available evidence places it: in the designed repertoire and its parameterisation. ^{7 5}

The collaboration joins synthesis and display without making them interchangeable. The same sequence can be represented through bits, byte values, a waveform or points. A viewer therefore encounters a designed interpretation of a numerical stream, not transparent access to the stream itself. This is a useful way to understand the project within generative art: selecting how a process becomes visible is as consequential as selecting the process. ^{6 5}

How the work was constructed

The program splits the token hash into pairs of hexadecimal characters, converts them into numbers and consumes values from the resulting array. Its selection function takes values from the end of that array and divides them by 256. These values govern choices within the supplied repertoire. They do not instruct the work to consult new randomness every time it opens. The token input fixes the configuration from which playback can proceed. ⁵

The six selectable families are weighted unequally. Melodic receives weight eight, heavy melodic six, and wobbly, power climb, absurdities and permutations three each. Another defined branch has zero selection weight. A function's presence in the source is consequently insufficient evidence that an edition can reach it. Similarly, the visual selection list includes the byte display twice, alongside one entry each for bits, waveform and starry. Those are implemented selection weights, not an audited census of realised feature frequencies. ⁵

Each family assembles a compact expression from arithmetic and bitwise operations. The families combine multiplication, division, shifts, masks and other transformations differently. Power climb introduces exponentiation; other branches use shifting and masking to produce changing numerical relationships. The assembled expression becomes a function of a running time index. Calling it a stored song would obscure the mechanism: the program computes successive values rather than playing a supplied sound recording. ⁵

The chosen synthesis rate is also structured. A hash-derived index enters a power-of-two calculation based on a twelve-step division, and the result is rounded. That rate is distinct from the audio context's requested 44,100 samples per second. During processing, the program maps its advancing counter into the synthesis function's rate. It masks the result to eight bits, recentres it around 128, divides by 256 and routes the signal through a gain value of 0.25. These are inspected numerical operations, not measurements of perceived loudness or pitch. ⁵

The audio implementation uses a ScriptProcessor with buffers of 1,024 samples. Its callback advances the counter while playback is active and supplies silence otherwise. The visual update is called from this same processing path. There is no separate animation clock in the inspected routine that can simply be assumed to continue when audio processing stops. MDN now documents ScriptProcessorNode as deprecated and identifies AudioWorklet as its replacement. That is a preservation concern for this particular implementation, not proof that every present-day browser fails to run it. ^{5 8}

The four display modes translate values differently. The bytes mode derives brightness from distance from the midpoint of the eight-bit range. The bits mode makes binary components visible as a sharply divided field. The waveform joins amplitude values, while the starry mode plots points. Internal canvas dimensions and sampling intervals differ between modes. A feature label therefore identifies a materially different mapping, rather than a decorative treatment applied after a common image has been completed. ⁵

The initial frame is itself selected. Before playback, the program tests one hundred sample positions, spaced one thousand units apart, and retains the candidate with the largest sum of pixel-channel values. The variable name describes this as density, but the implemented test is that numerical sum. The displayed opening frame is not necessarily the function at time zero. Close readings of a preview must acknowledge this small internal act of selection. ⁵

Interaction also requires careful reading. The source declares its activation function twice. The later declaration sets playback active and resumes the audio context; it replaces an earlier version that toggles a state. Describing a reliable click-to-pause control from the first declaration alone would misreport the deployed code. Pointer, keyboard and touch handlers supply activation routes. In a subsequent Chrome check, activation produced changing bands in #0 and an advancing audio-context clock. The collection-held #472 likewise changed from its opening grid into vertical strips and shifting checker-like regions after a click. Its audio clock could advance before that click while the picture remained stationary, so a running context alone does not prove that the composition has started. ^{5 1 9}

The mint

The platform identifies 2 April 2021 as the release date. Its index records the first mint at 05:47:06 that day, the second mint and scheduled start at 13:58:52, and activation at 14:04:24. The order is significant: the recorded second mint precedes the activation field. These values should remain separately named events rather than being compressed into one supposedly definitive launch time. Full timestamps are retained in the evidence record. ^{6 1 2}

The observed index reports 512 invocations against a maximum of 512, with completion recorded on 26 July 2021 at 18:52:47. Its price setting is 0.24 ETH. That current configuration does not independently establish the public opening price, a price paid by every recipient, or an uninterrupted sale at a single price. Recovering those histories requires contemporary announcements or transaction-level analysis beyond this inspection. ⁶

For the same reason, the interval between the earliest token and completion is not presented as a measured public sellout duration. The evidence does not classify artist allocations, administrative transactions and public purchases. The chronology supports an account of recorded mint activity; it does not justify supplying those missing distinctions by assumption. ⁶

Copyright and licensing

The indexed artwork licence field reads NIFTY, with no operative terms URL supplied. This essay records that exact label and classifies its meaning as unresolved. A name in a metadata field cannot supply the missing text, version or historical conditions. It should not be silently expanded into a Creative Commons licence or an unrestricted grant to reproduce or adapt the work. ⁶

No separate artist-code reuse permission was identified in the inspected project material. The script is analysed here without redistributing it as part of the essay. The browser's audio interface is a dependency of execution, but documentation for that interface is not a licence for the collaborators' program. AB5D's CC0 dedication applies to this original prose; it does not resolve those separate artwork and software permissions. ^{5 8}

Art Blocks’ June 2021 licensing guide identifies NIFTY as shorthand for NFT License 2.0. This strengthens the version attribution, although project-specific adoption and historical changes remain unverified.¹⁰

Close reading of four outputs

The sample comprises editions 0, 1, 10 and 20, with full token identifiers 38000000, 38000001, 38000010 and 38000020 on Ethereum contract 0xa7d8d9ef8d8ce8992df33d8b8cf4aebabd5bd270. These were selected as a small early-edition comparison. The observations concern initial canonical-renderer frames, not a survey of their changing states or a representative sample of all 512 works. ^{1 2 3 4}

In #0, an intense purple field is interrupted by black horizontal bands. Fine strips on the left give way to wider intervals toward the right. The image presents repetition at several scales: the eye can register a large block, then discover that it is composed of smaller alternating divisions. Its Bits display makes the change between values categorical. Dark and coloured areas meet abruptly, so the picture reads less as shading than as a set of explicit switches. ¹

This categorical appearance is also a trap for interpretation. The bands can resemble an enlarged notation, but their widths and positions should not be translated into a claimed heard rhythm without listening evidence. What can be said from the frame is that local repetition and larger interruption coexist. The image offers a legible visual structure while withholding a simple account of the duration that generated it. ^{1 5}

In #1, pink marks form a dense body near the bottom of the field, with high vertical spikes toward the left and an ascending diagonal array toward the right. Black gaps separate those concentrations. The Waveform mode changes the relation between figure and ground: instead of reading filled binary divisions, the eye follows connected excursions. A line that appears continuous at one scale becomes a mass of closely packed movements at another. This reading concerns its visible geometry; the Wobbly family name is not a substitute for an auditory description. ²

In #10, orange and black form repeated horizontal registers. Coarse dark blocks occupy the upper portion of each register, while finer patterns and quieter orange intervals extend below. Several subdivisions change near the middle of the image, creating a relationship between paired regions without making them identical. The Bytes mode permits intermediate intensities, so a dark shape can give way to an orange texture rather than simply disappear. Compared with #0, the frame feels layered and granular, although both arise from numerical values. ³

In #20, yellow and black return to the hard divisions of Bits. Short narrow stripes at the left repeat again near the centre, while larger black and yellow blocks occupy the right. The repeated beginnings encourage the eye to expect equivalence, but the larger intervals repeatedly interrupt it. The frame's force comes from that uneven distribution of scale. Its Goldenrod label does not do the descriptive work: the important relationship is between small alternating marks and the broad areas into which they expand. ⁴

Significance and limitations

ByteBeats makes representation part of its subject. A compact expression can supply a stream of numbers, but a stream is not yet a complete viewing experience. Choices of sampling, binary decomposition, brightness and line determine what becomes apparent. The inspected frames demonstrate that those choices can yield markedly different visual structures without severing the connection to computation. The project's economy lies in the relationship between rule and display, rather than in a claim that its entire technical apparatus fits into a few bytes. ^{5 1 2 3 4}

The largest remaining limit is the absence of a listening review. Visual playback and an advancing audio-context clock were observed for two outputs, but those observations cannot establish the audible character, fidelity or continuity of the sound. Nor can four opening frames and two playback checks establish the range of later states across the edition. The distinction keeps the account proportionate to what was actually inspected. ⁵

Unresolved research questions:

• Complete a listening review. Embedded visual playback has been checked; the original click-or-touch activation is preserved and disclosed beside the artwork.
• Inspect additional temporal states and the Starry mode, absent from this four-work sample.
• Recover launch-era price, edition announcements and any subsequent configuration changes.
• Classify mint transactions before calculating public-sale duration.
• Locate operative NIFTY terms and any separate artist-code reuse grant.
• Compare the indexed script with a pinned chain retrieval and document a reproducible preservation environment.