diff --git a/comfy_cli/command/knowledge.py b/comfy_cli/command/knowledge.py
index f5504b50c..bd5a5a504 100644
--- a/comfy_cli/command/knowledge.py
+++ b/comfy_cli/command/knowledge.py
@@ -231,14 +231,23 @@ def resolve_cmd(
     if model_id is None:
         knowledge.log_query("knowledge resolve", logged, hit_ids=[], zero_hit=True, bundle=bundle)
         q = query.strip().lower()
+        candidates = knowledge.model_candidates(bundle, query)
+        hint = "run `comfy knowledge status` to confirm a bundle is loaded; try a different alias"
+        details = {
+            "query": query,
+            "close_matches": difflib.get_close_matches(q, list(bundle.aliases), n=5, cutoff=0.6),
+        }
+        if candidates:
+            details["candidate_ids"] = candidates
+            hint = (
+                f"possible model ids (up to 5): {', '.join(candidates)}; "
+                "ask which model is intended, then retry `comfy knowledge resolve <exact-id>`"
+            )
         renderer.error(
             code="knowledge_unknown_model",
             message=f"no knowledge row matches {query!r}",
-            hint="run `comfy knowledge status` to confirm a bundle is loaded; try a different alias",
-            details={
-                "query": query,
-                "close_matches": difflib.get_close_matches(q, list(bundle.aliases), n=5, cutoff=0.6),
-            },
+            hint=hint,
+            details=details,
         )
         raise typer.Exit(code=1)
     knowledge.log_query("knowledge resolve", logged, hit_ids=[model_id], zero_hit=False, bundle=bundle)
diff --git a/comfy_cli/knowledge.py b/comfy_cli/knowledge.py
index aa65371a5..ab2864cad 100644
--- a/comfy_cli/knowledge.py
+++ b/comfy_cli/knowledge.py
@@ -145,6 +145,9 @@ class Bundle:
     # :func:`_resolve_tokens`.
     capability_tokens: tuple[tuple[tuple[tuple[str, int], ...], str, str], ...] = ()
     capability_context: dict[str, frozenset[str]] = field(default_factory=dict)
+    # Lowercase lookup keys whose raw spelling is eligible for suggestions.
+    # Kept separate because lower() can hide non-ASCII alphanumerics (e.g. K).
+    candidate_aliases: dict[str, str] = field(default_factory=dict)
 
 
 # One-element tuple so a memoized ``None`` is distinguishable from "not loaded yet".
@@ -339,6 +342,40 @@ def resolve(bundle: Bundle, query: str) -> dict | None:
     return bundle.models[model_id] if model_id is not None else None
 
 
+def _candidate_tokens(value: str) -> list[str]:
+    # Suggestions support ASCII names only. Never discard a Unicode qualifier
+    # or digit, or normalize it into a different family/version spelling.
+    if any(char.isalnum() and not char.isascii() for char in value):
+        return []
+    return re.findall(r"[a-z]+|[0-9]+", value.lower())
+
+
+def model_candidates(bundle: Bundle, query: str) -> list[str]:
+    """At most five canonical ids for an incomplete family/version name.
+
+    Suggestions only: even one candidate requires the caller to confirm the
+    model. Match the family word exactly, preserve every numeric component,
+    and allow omitted words, not substituted words or reordered tokens.
+    """
+    if len(query) > MAX_QUERY_CHARS:
+        return []
+    parts = _candidate_tokens(query)
+    numbers = [part for part in parts if part.isdigit()]
+    if not parts or not parts[0].isalpha() or not numbers:
+        return []
+    candidates: set[str] = set()
+    for alias, model_id in bundle.candidate_aliases.items():
+        if model_id not in bundle.models:
+            continue
+        words = _candidate_tokens(alias)
+        if not words or words[0] != parts[0] or [word for word in words if word.isdigit()] != numbers:
+            continue
+        remaining = iter(words[1:])
+        if all(part in remaining for part in parts[1:]):
+            candidates.add(model_id)
+    return sorted(candidates)[:5]
+
+
 def pick(bundle: Bundle, capability: str) -> dict | None:
     """Ranked picks for a capability, keyed exactly, then by spelling, then by wording.
 
@@ -650,16 +687,26 @@ def _index(data: dict, manifest: dict | None, *, source: str, stale: bool, path:
     # Exact ids first so no alias can shadow a model's own id; then the compiled
     # alias map, which wins over row-level aliases.
     aliases: dict[str, str] = {}
+    candidate_aliases: dict[str, str] = {}
+
+    def index_alias(alias: str, mid: str) -> None:
+        key = alias.lower()
+        owner = aliases.setdefault(key, mid)
+        # Check before casing loses information, and retain resolution precedence.
+        # A valid ASCII spelling may still support the same winning target.
+        if owner == mid and _candidate_tokens(alias):
+            candidate_aliases.setdefault(key, mid)
+
     for mid in models:
-        aliases.setdefault(mid.lower(), mid)
+        index_alias(mid, mid)
     compiled = data.get("aliases")
     if isinstance(compiled, dict):
         for alias, mid in compiled.items():
             if isinstance(alias, str) and isinstance(mid, str):
-                aliases.setdefault(alias.lower(), mid)
+                index_alias(alias, mid)
     for mid, row in models.items():
         for alias in _str_list(row.get("aliases")):
-            aliases.setdefault(alias.lower(), mid)
+            index_alias(alias, mid)
 
     caps_raw = data.get("capabilities")
     capabilities = (
@@ -739,6 +786,7 @@ def _index(data: dict, manifest: dict | None, *, source: str, stale: bool, path:
         capabilities=capabilities,
         deprecations=deprecations,
         aliases=aliases,
+        candidate_aliases=candidate_aliases,
         templates=templates,
         nodes=nodes,
         model_capabilities=model_capabilities,
diff --git a/docs/json-output.md b/docs/json-output.md
index 6e39597f4..9a2c493cd 100644
--- a/docs/json-output.md
+++ b/docs/json-output.md
@@ -777,6 +777,24 @@ are objects in the normal case; treat a non-object item as an opaque message.
 `node_id` is always taken from the payload's own map key, so a server-supplied
 `node_id` field inside a record cannot override it.
 
+### Incomplete model names
+
+`comfy knowledge resolve` resolves declared model ids and aliases. An incomplete
+family/version name such as `Qwen 2.1` still returns `knowledge_unknown_model`
+with exit code 1. When catalog ids or aliases contain the same family word and
+numeric components, in order, `error.details.candidate_ids` lists up to five
+sorted, distinct canonical ids. Omitted words are allowed for suggestions;
+substituted family words and different versions are not.
+Suggestions conservatively skip queries and catalog aliases containing non-ASCII
+letters or numbers rather than discard or normalize those qualifiers. Declared
+id and alias resolution still uses its existing rules.
+
+These are suggestions, even when there is only one. Ask which model is intended,
+then retry `comfy knowledge resolve <exact-id>`. They grant no permission to
+download or run a model. Successful full-name resolution and unknown-name errors
+without candidates retain their existing behavior; `close_matches` remains a
+separate spelling hint.
+
 ## Output object
 
 Entries of `executed.outputs`:
diff --git a/tests/comfy_cli/test_knowledge.py b/tests/comfy_cli/test_knowledge.py
index 58f538fe1..3e8214ce6 100644
--- a/tests/comfy_cli/test_knowledge.py
+++ b/tests/comfy_cli/test_knowledge.py
@@ -1506,3 +1506,205 @@ def test_the_bare_listing_asks_nothing_so_logs_nothing(self, tmp_path, monkeypat
         _env_bundle(tmp_path, monkeypatch)
         _run(["pick"], capsys)
         assert queries == []
+
+
+class TestNamedModelDisambiguation:
+    """Missing family words warrant a choice, never an inferred model."""
+
+    @pytest.fixture
+    def named_bundle(self, monkeypatch):
+        # Identity/aliases from comfy-knowledge v0.2.2 (7ee63b21a59b),
+        # canon/models/image/qwen-image-2-1.yaml. Cloud's compiled bundle is
+        # sha256 53a70dd644dbca30ad7c20b7f8bd8c00d154609ae1edf22e63aa834954f11d48.
+        data = {
+            "models": {
+                "qwen-image-2-1": {
+                    "id": "qwen-image-2-1",
+                    "aliases": [
+                        "Qwen Image 2.1",
+                        "Qwen-Image-2.1",
+                        "Qwen Image 2.1 T2I",
+                        "Qwen Image 2.1 Edit",
+                        "Qwen Image 2.1 Remove Background",
+                        "Qwen Image 2.1 Turbo",
+                        "Qwen-Image-2.1-Turbo",
+                    ],
+                },
+                "qwen-image-2512": {"id": "qwen-image-2512"},
+                "qwen-image-edit": {"id": "qwen-image-edit"},
+            }
+        }
+        bundle = knowledge._index(data, None, source="env", stale=False, path="fixture", mtime=0.0)
+        monkeypatch.setattr(knowledge, "load_bundle", lambda **kw: bundle)
+        return bundle
+
+    @pytest.mark.parametrize("query", ["Qwen 2.1", "QWEN-2-1", "qwen2.1"])
+    def test_short_name_offers_canonical_id_without_resolving(self, named_bundle, query, capsys):
+        assert knowledge.resolve_id(named_bundle, query) is None
+        rc, env = _run(["resolve", query], capsys)
+        assert rc == 1
+        assert env["ok"] is False
+        assert env["data"] is None
+        assert env["error"]["code"] == "knowledge_unknown_model"
+        assert env["error"]["details"]["candidate_ids"] == ["qwen-image-2-1"]
+        assert "ask which model" in env["error"]["hint"]
+        assert "qwen-image-2-1" in env["error"]["hint"]
+
+    @pytest.mark.parametrize("query", ["Qwen Image 2.1", "qwen-image-2-1", "QWEN IMAGE 2.1 TURBO"])
+    def test_full_name_still_resolves(self, named_bundle, query, capsys):
+        rc, env = _run(["resolve", query], capsys)
+        assert rc == 0
+        assert env["data"]["id"] == "qwen-image-2-1"
+        _validate(env["data"])
+
+    @pytest.mark.parametrize(
+        "query",
+        [
+            "Qwen VL 2.1",
+            "Qwen text 2.1",
+            "Qwen 2.10",
+            "Qwen 21",
+            "Qwen 1.2",
+            "Qwen 2",
+            "Qwen",
+            "other 2.1",
+            "unknown-pm2214",
+            "",
+            "Qwen" + " " * knowledge.MAX_QUERY_CHARS + "2.1",
+            "Qwen 文本 2.1",
+            "Qwen 2.1β",
+            "Qwen 2.1２",
+            "Qwen 2.1K",
+            "Qwen 2.1İ",
+            "Qwеn 2.1",  # Cyrillic e
+        ],
+    )
+    def test_unknown_or_different_family_version_stays_unknown(self, named_bundle, query, capsys):
+        rc, env = _run(["resolve", query], capsys)
+        assert rc == 1
+        assert env["error"]["code"] == "knowledge_unknown_model"
+        assert env["error"]["details"].get("candidate_ids", []) == []
+        assert "ask which model" not in env["error"]["hint"]
+
+    @pytest.mark.parametrize(
+        "alias",
+        [
+            "Qwen Image 文本 2.1",
+            "Qwen Image 2.1β",
+            "Qwen Image 2.1２",
+            "Qwen Image 2.1²",
+            "Qwen Image 2.1K",
+            "Qwen Image 2.1İ",
+            "Qwеn Image 2.1",
+        ],
+    )
+    @pytest.mark.parametrize("top_level", [False, True])
+    def test_unicode_aliases_cannot_offer_ascii_candidates(self, monkeypatch, capsys, alias, top_level):
+        data = {"models": {"opaque-model": {"aliases": [] if top_level else [alias]}}}
+        if top_level:
+            data["aliases"] = {alias: "opaque-model"}
+        bundle = knowledge._index(data, None, source="env", stale=False, path="fixture", mtime=0.0)
+        monkeypatch.setattr(knowledge, "load_bundle", lambda **kw: bundle)
+        rc, env = _run(["resolve", "Qwen 2.1"], capsys)
+        assert rc == 1
+        assert env["error"]["code"] == "knowledge_unknown_model"
+        assert env["error"]["details"].get("candidate_ids", []) == []
+        assert "ask which model" not in env["error"]["hint"]
+        # Suggestion eligibility must not change declared alias resolution.
+        for spelling in (alias, alias.lower(), alias.upper()):
+            assert knowledge.resolve_id(bundle, spelling) == "opaque-model"
+            rc, env = _run(["resolve", spelling], capsys)
+            assert rc == 0
+            assert env["data"]["id"] == "opaque-model"
+
+    @pytest.mark.parametrize("top_level", [False, True])
+    @pytest.mark.parametrize("reverse", [False, True])
+    def test_candidate_alias_collisions_preserve_resolution_precedence(self, top_level, reverse):
+        pairs = [("Qwen Image 2.1K", "opaque-unicode"), ("Qwen Image 2.1k", "opaque-ascii")]
+        if reverse:
+            pairs.reverse()
+        data = {"models": {mid: {"aliases": [] if top_level else [alias]} for alias, mid in pairs}}
+        if top_level:
+            data["aliases"] = dict(pairs)
+        b = knowledge._index(data, None, source="env", stale=False, path="fixture", mtime=0)
+        assert knowledge.resolve_id(b, "Qwen Image 2.1k") == pairs[0][1]
+        assert knowledge.model_candidates(b, "Qwen 2.1") == (["opaque-ascii"] if reverse else [])
+
+    @pytest.mark.parametrize("reverse", [False, True])
+    def test_eligible_spelling_for_same_target_still_offers_candidate(self, reverse):
+        aliases = ["Qwen Image 2.1K", "Qwen Image 2.1k"]
+        if reverse:
+            aliases.reverse()
+        b = knowledge._index(
+            {"models": {"opaque-model": {"aliases": aliases}}},
+            None,
+            source="env",
+            stale=False,
+            path="fixture",
+            mtime=0,
+        )
+        assert knowledge.model_candidates(b, "Qwen 2.1") == ["opaque-model"]
+        assert knowledge.resolve_id(b, aliases[0]) == "opaque-model"
+
+    def test_canonical_and_top_level_aliases_keep_precedence(self):
+        b = knowledge._index(
+            {
+                "models": {
+                    "qwen-image-2-1": {},
+                    "opaque-model": {"aliases": ["qwen-image-2-1", "Qwen Text 2.1"]},
+                },
+                "aliases": {"qwen-image-2-1": "opaque-model", "Qwen Text 2.1": "qwen-image-2-1"},
+            },
+            None,
+            source="env",
+            stale=False,
+            path="fixture",
+            mtime=0,
+        )
+        assert knowledge.resolve_id(b, "qwen-image-2-1") == "qwen-image-2-1"
+        assert knowledge.resolve_id(b, "Qwen Text 2.1") == "qwen-image-2-1"
+        assert knowledge.model_candidates(b, "Qwen 2.1") == ["qwen-image-2-1"]
+
+    def test_raw_unicode_model_id_cannot_offer_ascii_candidates(self):
+        mid = "Qwen Image 2.1K"
+        b = knowledge._index({"models": {mid: {}}}, None, source="env", stale=False, path="fixture", mtime=0)
+        assert knowledge.resolve_id(b, mid) == mid
+        assert knowledge.model_candidates(b, "Qwen 2.1") == []
+
+    def test_future_image_vl_and_text_models_remain_an_explicit_choice(self, monkeypatch, capsys):
+        # Synthetic future catalog: no preference for Image, VL, or text.
+        ids = ["qwen-image-2-1", "qwen-vl-2-1", "qwen-text-2-1"]
+        b = knowledge._index(
+            {"models": {mid: {"id": mid} for mid in ids}},
+            None,
+            source="env",
+            stale=False,
+            path="fixture",
+            mtime=0.0,
+        )
+        monkeypatch.setattr(knowledge, "load_bundle", lambda **kw: b)
+        rc, env = _run(["resolve", "Qwen 2.1"], capsys)
+        assert rc == 1
+        assert env["error"]["details"]["candidate_ids"] == sorted(ids)
+        assert "ask which model" in env["error"]["hint"]
+        assert knowledge.resolve_id(b, "Qwen 2.1") is None
+        assert knowledge.model_candidates(b, "Qwen VL 2.1") == ["qwen-vl-2-1"]
+
+    def test_candidates_are_general_sorted_bounded_and_deduplicated(self):
+        ids = [f"acme-variant{letter}-2-1" for letter in "hgfedcba"]
+        b = knowledge._index(
+            {
+                "models": {mid: {"id": mid, "aliases": [mid.upper(), mid.replace("-", " ") + " turbo"]} for mid in ids},
+                "aliases": {"Acme Missing 2.1": "missing-row"},
+            },
+            None,
+            source="env",
+            stale=False,
+            path="fixture",
+            mtime=0.0,
+        )
+        assert knowledge.model_candidates(b, "Acme 2.1") == sorted(ids)[:5]
+        boundary_query = "Acme 2.1".ljust(knowledge.MAX_QUERY_CHARS)
+        assert len(boundary_query) == 200
+        assert knowledge.model_candidates(b, boundary_query) == sorted(ids)[:5]
+        assert knowledge.model_candidates(b, boundary_query + " ") == []