civarchive.com

What is CivArchive (CivitAI Archive)?

CivArchive (CivitAI Archive) is a community-driven search and archive site for AI models, formerly called CivitAI Archive. Its core pitch is cross-platform discovery: you search, verify and track models using SHA256 hashing, and it points to listings on CivitAI, HuggingFace, TensorArt, SeaArt, PixAI, Shakker, Tungsten, Civision, TensorHub, Yodayo, Moescape and torrents. Rather than hosting one model library, it acts as an index over many.

Its homepage is organized around CivitAI-centric rankings: top "safe" models from the past week, month and quarter, plus latest safe models and recently deleted safe models. Listings show model type (LORA, Checkpoint, VAE, TextEncoder, Workflows, Other), a base-model family such as Qwen 2.1, MiniMax H3 or Krea 2, the creator name, and a download or view count. The "recently deleted" section is the distinctive part: it is aimed at people who need to find a model after its original page disappears.

Who it is for

• Model hunters who saw a model on one platform and want to find it again elsewhere.
• ComfyUI and workflow users looking for base-model-compatible LORAs, VAEs, text encoders and ready-made workflows.
• Archivists and researchers tracking which models exist, which are popular, and what has been removed.
• Cautious downloaders who want a hash to compare against a file they already have.

How to use it in practice

1. Search by model name, creator or hash.
2. Check the model type and base model before downloading, since a Qwen 2.1 LORA will not work with a different architecture.
3. Use the SHA256 hash to confirm a file matches the listing, especially when the file came from a mirror or torrent.
4. If a model has vanished from its original host, check the recently deleted list.

Trade-offs to keep in mind

The value here is breadth and persistence, not curation or support. Rankings are labeled "safe," which reflects the site's own filtering rather than a guarantee about content or licensing. Download counts are popularity signals, not quality signals. Because files may live on third-party hosts or torrents, hash verification is the practical safeguard. And as an archive, it can lag behind or miss models that were never indexed.

A useful next step: pick one model you already have, search for it here, and compare the listed SHA256 with your local file. That single test tells you whether this archive is reliable enough for your workflow.

How do I use SHA256 hashing on CivArchive to verify and track an AI model across platforms?

SHA256 hashing on CivArchive works as a fingerprint system: you take the hash of a model file you already have, then use it to confirm whether the same file exists in the archive and where else it has appeared. It is most useful when a model's name, version label or page has changed, or when a listing has been removed and you want to know whether your copy is the genuine article.

How the workflow fits together

1. Get the hash of your local file. On most systems this is a single command-line step (sha256sum on Linux, shasum -a 256 on macOS, certutil -hashfile on Windows). The output is a 64-character string.
2. Search that string on CivArchive. The site's stated purpose is to search, verify and track AI models using SHA256 hashing, so the hash is the lookup key rather than the model name.
3. Read the result as an identity check. A match tells you the file you hold corresponds to a known archived entry. No match can mean a re-export, a re-quantised version, a renamed repack, or simply a file the archive has not indexed.
4. Follow the cross-platform trail. CivArchive says it covers CivitAI, HuggingFace, TensorArt, SeaArt, PixAI, Shakker, Tungsten, Civision, TensorHub, Yodayo, Moescape and torrents, so one hash can point you to mirrors of the same artifact on other hosts.

What the archive's own listings show

The page surfaces ranked and dated collections: top safe models over the past week, month and quarter, latest safe models, and recently deleted safe models. The deleted-models view is the practically important one — if a creator pulls a file or a platform removes it, the archived hash remains a way to identify what you have and to judge whether a file circulating elsewhere is the same artifact.

A realistic scenario

You downloaded a LoRA months ago and the CivitAI page is now gone. You have the file but no page, no version number and no author confirmation. Hashing the file and searching that hash on CivArchive can tell you whether it matches an archived record, which model family it belongs to, and whether the same hash appears on another platform. That is verification of provenance, not a guarantee of safety or licensing.

Practical decision criteria

• Trust the hash, not the filename. Renamed files are common; identical hashes are not.
• Treat a mismatch as a question, not a verdict. Quantised, pruned or re-encoded variants legitimately produce different hashes.
• Use the archive for tracking, not for endorsement. A listing confirms existence and identity, not quality or permissions.
• Check the deleted list before assuming a model is gone forever. It is often the fastest route to identifying an orphaned file.

For a broader reference on the same model, the original host is CivitAI, and Hugging Face hosts many mirrors under Hugging Face.

How can I find a CivitAI model that was recently deleted?

Use CivArchive's "Recently deleted safe models" section, then confirm the file by its SHA256 hash rather than by name alone. Deleted CivitAI models are often re-uploaded, renamed, or mirrored, so the hash is the reliable identifier.

On CivArchive (CivitAI Archive), the page groups models into weekly, monthly, and quarterly "safe" lists plus a "Recently deleted safe models" view. Start there if you remember roughly when the model disappeared. If you know the exact filename or creator, search that instead — the archive indexes model pages and files across multiple platforms and torrents, so a match may surface even when the original CivitAI page is gone.

A practical workflow

1. Note everything you remember: model name, creator, base model (for example SDXL, Pony, Illustrious, or a Qwen/Flux variant), approximate upload date, and any preview image.
2. Search CivArchive by name and by creator, then open the model page and copy its SHA256 hash.
3. Search that hash across the other platforms CivArchive lists. A matching hash means the same file, regardless of the title it now carries.
4. If the hash appears only in a torrent listing, treat the download as unverified until you check the file yourself.

Why the hash matters

A concrete scenario

Suppose a LoRA you used last year no longer loads from its CivitAI link. You remember it was a detail-enhancer LoRA for a Qwen-based image model and the creator's handle. Search the handle on CivArchive, filter mentally by the Qwen family, open the closest match, and compare its hash with the copy still sitting in your local models/Lora folder. If the hashes match, you have the original file and only need to restore the metadata sidecar. If they do not, keep looking rather than assuming the re-upload is identical.

Limits worth knowing

The archive is community-driven, so coverage is uneven: popular and recent models are more likely to be captured than obscure ones, and "safe" labeling is a filter, not a guarantee about content. A deleted model may also have been removed for licensing or policy reasons, so check the creator's terms before redistributing anything you recover.

How do I copy a model from CivitAI to HuggingFace or another platform using CivArchive?

CivArchive is a search-and-verification index, not a migration tool. It helps you find a model across platforms and confirm you have the same file, but it does not move files between CivitAI and HuggingFace for you. The "copy" is a manual download-and-reupload process that CivArchive makes safer by giving you a common fingerprint.

What CivArchive actually gives you

The site describes itself as a way to "search, verify, and track AI models using SHA256 hashing" across CivitAI, HuggingFace, TensorArt, SeaArt, PixAI, Shakker, Tungsten, Civision, TensorHub, Yodayo, Moescape and torrents. In practice that means:

• A cross-platform search index so you can check whether a model already exists elsewhere before re-uploading it.
• SHA256 hashes so you can confirm two files are byte-identical rather than trusting a filename.
• Lists such as top safe models by week, month and quarter, latest safe models, and recently deleted safe models — useful when a CivitAI page disappears and you need to know what the file was.

The practical workflow

1. Find the model on CivArchive and note its SHA256 hash and exact version.
2. Download the file from the original platform (CivitAI, or wherever CivArchive points).
3. Compute the hash locally and compare it to the listed value.
4. If you are uploading to HuggingFace or another host, check CivArchive first for an existing copy — re-uploading a duplicate is the most common mistake.
5. When you re-upload, keep the creator's name, model version and license terms intact.

Why the hash step matters

Filenames and version labels get reused constantly. Two files called "v1.0" can differ, and a re-upload can silently be a pruned, quantized or otherwise altered build. A matching SHA256 is the only reliable way to say "this is the same file." For a model like a Qwen Image 2.1 checkpoint, where versions and formats (diffusers, VAE, text encoder) are distributed separately, that distinction decides whether your results will match someone else's.

Trade-offs and limits

CivArchive indexes and verifies; it does not host, mirror or transfer files, and it cannot grant you redistribution rights. "Safe" labels in its lists reflect that site's own filtering, not a legal clearance. If the original model page is gone, you may find the hash but still have no licensed source to download from — in that case, treat the model as unavailable rather than pulling it from an unverified mirror.

A concrete scenario

You want a LoRA you saw on CivitAI to also live in your HuggingFace repo for a team workflow. Look it up on CivArchive, record the SHA256, download the original, verify the hash, confirm no copy already exists on HuggingFace, then re-upload with attribution and the original license. The hash is what lets your teammates confirm they pulled the same weights you tested.

For broader model discovery beyond CivitAI's ecosystem, Hugging Face and CivitAI are the two primary sources CivArchive points back to.

What are the top safe models on CivitAI this week or month?

CivArchive's own leaderboards list "safe" models by period, so the answer depends on which window you check. This week's top entries are dominated by Qwen 2.1 tooling: LoRAs such as PornMaster Qwen Image 2.1 Age Slider V1 and Elusarca's Qwen 2.1 Detail Enhancer, plus workflows like Qwen 2.1 2K Upscale and a character reference sheet generator. Over the past month, Krea 2 and MiniMax H3 items join the Qwen crowd — for example Krea2 TextFusion Refusal-Reduction LoRA and MiniMax H3 Turbo.

What "safe" means here

It is a filter label on the archive, not a guarantee about content or licensing. Treat it as "surfaced without the flagged-content tag," and still read each model's own page before downloading.

Week vs month at a glance

A practical next step

If you want a stable pick, start with the monthly list and check download counts and update dates on the model page; use the weekly list when you specifically want the newest Qwen 2.1 or MiniMax H3 tooling. CivArchive also tracks the same models across Hugging Face and other hosts via SHA256 hashes, which helps confirm you have the exact file a workflow expects.

How does CivArchive differ from CivitAI for browsing and downloading models?

CivArchive is a third-party archive and search layer for AI models, not a primary host. CivitAI is where models are originally uploaded, versioned and discussed. CivArchive mirrors or indexes those files, adds cross-platform search, and keeps records of models that have since disappeared from their original host. For browsing and downloading, the practical difference is: CivitAI is the living source of truth; CivArchive is a durable index and fallback.

What each site is good at

CivitAI is the better first stop when you want the current version, creator description, comments, reviews, example images and the normal download flow. If a model is still live, downloading from the source is usually simplest and most likely to match what the creator intended.

CivArchive is more useful when the source page is gone, when you want to verify a file's identity, or when you want to search across multiple platforms at once. Its page evidence shows SHA256 hashing as a core feature, plus lists such as "Top safe models in the past week," "Latest safe models" and "Recently deleted safe models." That last category is the clearest reason to use it: it tracks models that have been removed from CivitAI.

Browsing differences

CivitAI browsing is organised around creators, model pages, tags, reviews and community activity. CivArchive browsing is organised around archive status and discovery lists: trending, recent and deleted models, with model type, base model and file format shown in each entry.

Downloading and verification

On CivitAI, downloading is tied to the model page and its current version. On CivArchive, the emphasis is on locating a file and confirming it by hash, which matters if you are re-downloading something from a mirror, a torrent or another platform. If you already have a file, hashing it and searching CivArchive is a quick way to confirm what it is and whether it matches a known release.

Who should use which

Casual users who just want a current model with examples and comments should start on CivitAI. Archivists, researchers, and anyone rebuilding a workflow after a model disappears should use CivArchive. People who work across several model platforms will get the most from CivArchive's cross-platform search, while creators will still want CivitAI for publishing and feedback.

For related official sources, see CivitAI and Hugging Face.

Next step: if you are looking for a specific model, search its name on CivitAI first. If it is missing there, search CivArchive, and if you have the file already, run its SHA256 hash through CivArchive to confirm you have the right one.